What shipped, version by version.

0.41.0

• Added agentloop start as the repo-native preflight for software agents, with active task, decisive preflight state, read-first source handles, risk summary, next safe command, and context-budget impact.
• Added Start preflight states for ready continuation, missing task, missing or stale verification, scope drift, review-ready work, blocked risk, and evidence-only sessions.
• Added agentloop_start to the read-only MCP surface so agent platforms can request the same briefing without broad repo scans.
• Added Start guidance to generated Codex, Claude Code, Cursor, OpenCode, Gemini CLI, GitHub Copilot CLI, and generic agent instructions.
• Added Start-focused README, CLI reference, MCP, getting-started, and Context Contract docs plus a regenerated README demo GIF centered on the preflight flow.
• Added Start tests for JSON shape, Markdown output, goal-aware states, scope drift, risk blocking, MCP payloads, and docs drift.

0.40.0

• Added agentloop context budget, agentloop context pack, and agentloop context show as a unified Context Contract for software agents.
• Added auditable context-pack receipts that explain included evidence, omitted broad context, source handles, verification freshness, next actions, and context-budget estimates.
• Added read-only MCP tools for context budget, context pack, and context handle expansion so agent platforms can consume the same contract directly.
• Added generated agent guidance that tells Codex, Claude Code, Cursor, OpenCode, Gemini CLI, GitHub Copilot CLI, and generic agents to request a context pack before broad repo reads.
• Added public Context Contract documentation, CLI reference coverage, README positioning, an ASCII workflow diagram, a context-budget visual, and a regenerated terminal demo GIF.
• Added tests for context-pack JSON and Markdown output, redacted handle expansion, MCP tool exposure, CLI docs drift, and public documentation hygiene.

0.39.0

• Added agentloop guard for local drift, proof-debt, stale-verification, baseline, watch-mode, report-writing, and context-budget checks.
• Added deterministic evidence maps through agentloop explain-diff, with changed-file coverage against active task scope, recent run evidence, verification freshness, and risk-sensitive areas.
• Added agentloop resume-pack for compact continuation briefs tailored to Codex, Claude, Cursor, generic agents, and human reviewers.
• Added context-budget estimates to Guard, resume packs, and review context so agents can choose compact local evidence instead of pasting broad chat history or changed-file lists.
• Added Guard and evidence-map summaries into review-context, ship, and prepare-pr so normal review surfaces explain scope coverage and next actions.
• Added public Guard and evidence-map docs plus README visuals for the context-budget workflow and regenerated terminal demo.
• Changed doctor monorepo detection to catch nested package manifests such as apps/*/package.json, functions/package.json, and test package layouts when the root package does not declare workspaces.
• Changed AgentLoop evidence classification so .agentloop/state.json and repo-local task contracts are treated as local evidence instead of unexplained product work.
• Fixed real-repo trial friction around generated harness baselines by documenting when to commit the initialized AgentLoop files before feature work.

0.38.0

• Added additive status and next loop guidance so agents re-entering a task see the matching .agentloop/loops/<type>.md file when it exists.
• Added additive create-task loop guidance so typed tasks point to the matching .agentloop/loops/<type>.md file when it exists.
• Added a focused product-positioning test for internal product-direction files and aligned persona/panel wording with agent-assisted engineering language.
• Added a research task type and local research loop guidance for reviewable research plans, findings, limits, and follow-up tasks.
• Added --redact-paths support to task lifecycle commands so task set, task status, task done, task archive, and task clear can be used in shareable command logs.
• Changed create-task help and generated AGENTLOOP guidance to use agentic engineering-session wording.
• Added package metadata tests for unsupported positioning so npm description and keywords stay aligned with software-agent and agent-assisted engineering language.
• Changed public product wording and package metadata to use software-agent and agent-assisted engineering language, and added public-doc hygiene guards against cheap assistant-style positioning.
• Added create-task warnings for review-critical placeholder sections so draft task contracts stay permissive but agents see missing problem, outcome, scope, acceptance, verification, or rollback content immediately in human and JSON output.
• Added create-task warnings when a new task starts over existing dirty non-evidence files, with bounded human and JSON examples so agents can review scope before implementation.
• Updated generated task-contract guidance so freshly initialized repos explain dirty-work and placeholder-section create-task warnings without changing runtime behavior.
• Updated generated AgentFlight placeholder recovery guidance so root, harness, and agent instruction templates tell agents to run status plus task doctor, preserve placeholder evidence, clear placeholder active state, then pin or create a real task.
• Changed task clear output to distinguish cleared persisted active-task pointers from no-op runs, including ignored AgentFlight placeholder pointers, without deleting task files.
• Changed human-readable artifacts output to label archived task fallbacks as Latest archived task evidence while keeping JSON compatibility through archived: true.
• Changed human-readable check-gates output to label archived task fallbacks as Archived task evidence while keeping JSON gate fields stable.
• Changed check-gates create-task next-action guidance to include bounded dirty non-evidence file examples while keeping gate decisions unchanged.
• Changed human-readable review-context gate summaries to label task evidence as active, latest open, archived, or missing while keeping JSON gate fields stable.
• Changed human-readable ship reports to show an inherited dirty-work baseline section when task Risk Notes include the generated pre-existing dirty non-evidence count, while preserving readiness scoring and JSON output.
• Changed status and next create-task recommendations to mention existing dirty non-evidence files before agents start the next task.
• Changed dirty-work warning terminology from non-AgentLoop to non-evidence so local .agentloop product or harness files are described accurately when they are not generated evidence.
• Changed create-task so dirty-work warnings also persist a bounded Risk Notes bullet with the pre-existing dirty non-evidence count and examples.
• Added npm run typecheck to npm run maintenance:check so the recurring guard catches TypeScript regressions before later docs, safety, and dogfood checks pass.
• Changed agentloop ship reports to render task Risk Notes with Markdown-safe single-line prose and a clear fallback when no risk notes were recorded.
• Changed agentloop review-context to report the active task Risk Notes count without printing the task risk-note prose.
• Fixed readiness scoring so final task Markdown sections, including final Rollback Notes or Risk Notes, are parsed correctly while placeholder risk-note text remains ignored.
• Added tests/readiness-score.test.ts to npm run test:unit so the recurring maintenance guard exercises deterministic review-readiness scoring coverage.
• Changed dogfood gates to print AgentFlight session status after AgentFlight doctor so blocked session readiness is visible without parsing human output.
• Documented the safe AgentFlight verification syntax, agentflight verify -- <command> <args>, so agents do not pass full shell commands as one quoted argument.
• Changed human-readable maintainer-check output to label package-manifest-only and combined package/dependency warnings precisely while preserving the existing JSON check id.
• Changed agentloop maintainer-check to distinguish package manifest changes from dependency lockfile changes while preserving the existing warning check id.
• Clarified npm run dogfood:strict guidance so strict review-gate warnings are distinguished from maintainer-check reviewer guidance warnings.
• Changed prepare-pr and deterministic PR summaries to surface verification report Not Run entries directly, with a clear fallback when no commands were skipped.
• Changed the PR handoff and deterministic summary heading for those entries to Verification Report Not Run so reviewers see that the list comes from report evidence rather than a broad coverage claim.
• Changed prepare-pr PR bodies to collapse the optional imported GitHub metadata section cleanly when no metadata exists, avoiding an extra blank gap before the reviewer checklist.
• Changed verification report Markdown to show configured command strings beside Not Run aliases while preserving existing JSON notRun alias values.
• Changed verification reports to omit configured Not Run entries when an exact same command string ran from task verification.
• Changed dirty create-task next-action guidance to include bounded dirty non-evidence file examples while keeping the recommended command unchanged.
• Changed compact status --brief output to show verification=previous:<status> when no active or open task exists.
• Changed human review-context output to label retained latest verification as previous evidence when no active or open task exists.
• Changed human next output to label the latest verification as previous evidence when no active or open task exists, matching status while keeping JSON output unchanged.
• Changed human status output to label the latest verification as previous evidence when no active or open task exists, while keeping JSON latestReport unchanged.
• Changed agentloop artifacts --stale to show candidate counts grouped by artifact type before the bounded human preview, with deterministic JSON summary data.
• Fixed path redaction so external Markdown URLs are preserved while local root variants still redact fully in verification evidence.
• Changed maintainer-check and release-check path redaction to reuse the shared URL-safe local-root helper, including /var and /private/var path variants.
• Changed ship readiness scope warnings to include compact non-evidence review-area counts when a change set is broad.
• Changed ship report diff stats to include compact untracked non-evidence file markers while preserving Git's tracked-file stats and the full JSON changed-file inventory.
• Changed deterministic summaries, handoffs, and HTML reports to use the same untracked-aware diff-stat rendering as ship reports.

0.37.0

• Added agentloop doctor --advisory for onboarding and real-repo trial preflight checks that should show failing diagnostics without failing the shell.
• Changed agentloop status and agentloop next to route active task contracts with review-critical placeholder sections to agentloop task doctor before verification or handoff.

0.36.2

• Broke the artifacts/runs/task-state core import cycle by moving generated artifact patterns and output-path validation into a leaf module.
• Clarified direct AgentFlight placeholder recovery so task doctor tells agents to preserve placeholder files as session evidence while clearing or re-pinning real task state.
• Added a compact first-use loop to the README and getting-started docs.
• Documented why the intentional prepublishOnly release guard remains acceptable while other lifecycle scripts still require review.
• Added a real-repo trial checklist for policy packs and imported GitHub metadata before expanding bundled packs or scoring behavior.
• Added public-doc hygiene coverage that keeps real-repo trial guidance local-first, non-marketing, and scoring-neutral.

0.36.1

• Fixed Windows smoke fixtures and release-status report path normalization so cross-platform release checks compare stable, relative evidence paths.
• Fixed doctor --redact-paths coverage for Windows drive variants, current-directory variants, inferred AgentLoop artifact roots, and labeled local path roots.
• Clarified agentloop status --brief --json help copy so compact JSON output is described as machine-readable.
• Mirrored raw AgentFlight recovery guidance into generated agent instruction templates so direct AgentFlight sessions can re-pin detailed AgentLoop task contracts.

0.36.0

• Added stale task-state recovery across agentloop status, agentloop next, and agentloop task doctor so missing or stale .agentloop/state.json pointers are reported with bounded recovery guidance instead of broad scans or mutation.
• Added agentloop artifacts fallback support for the latest archived terminal task, including JSON metadata and human output that identify archived task evidence when no live task is available.
• Hardened release and review evidence surfaces by separating AgentLoop-generated evidence churn from source changes across status, next, check-gates, review-context, artifacts, runs, ship, maintainer-check, and release-check output.
• Improved AgentFlight placeholder handling so placeholder tasks are grouped, parked, and excluded from latest-task evidence where real task contracts should take precedence.
• Expanded built CLI smoke coverage for release-check evidence splitting, stale task-state recovery, archived task fallback, redaction, generated artifact ordering, policy packs, npm-status, and review surfaces.
• Added maintenance coverage for npm-status, SchemaStore, GitHub metadata, policy packs, public docs, links, ProjScan, AgentFlight, and strict dogfood checks.
• Tightened Markdown-safe and redacted CLI output across task, policy, install-agent, verify, handoff, report, badge, ci-summary, release-notes, release-proof, schemastore, run-ledger, status, next, check-gates, ship, prepare-pr, maintainer-check, and artifact commands.

0.35.2

• Added explicit author metadata to the root composite GitHub Action and aligned the Action metadata quoting with the Marketplace-published ProjScan Action.
• Prepared a Marketplace-focused patch release so GitHub evaluates the released Action metadata again.

0.35.1

• Added GitHub Marketplace proof to agentloop release-proof, including --only github-marketplace and captured --github-marketplace-json support.
• Added Marketplace-facing metadata to the root composite GitHub Action so the published Action has a badge, clearer description, and explicit trusted-input copy.
• Updated release docs and checklists so maintainers must publish the Action through the GitHub release UI Marketplace checkbox and verify the Marketplace listing before claiming cross-channel release proof.

0.35.0

• Added agentloop verify --post-verification-gates so reviewed task post-verification gates can run after the verification report exists while default verify behavior stays unchanged.
• Added focused GitHub metadata safety tests to npm run maintenance:check so the optional/read-only metadata contract is checked directly.
• Added focused policy-pack safety tests to npm run maintenance:check so local/no-overwrite policy-pack behavior is checked directly.
• Added release-proof HEAD/tag clarity so maintainers can see when public channels match the package version but the current checkout contains unreleased commits.
• Fixed generated artifact ordering so timestamped verification, handoff, and ship evidence stays ordered by filename even when Git operations rewrite filesystem mtimes.
• Added a dedicated scripts/maintenance-check.mjs gate behind npm run maintenance:check that exercises near-term roadmap health checks for release proof, public docs, SchemaStore, policy packs, read-only GitHub metadata, AgentFlight, ProjScan, and non-strict dogfood.
• Changed the dogfood ProjScan command to use the documented global --format flag position.

0.34.1

• Added release-delta diagnostics to agentloop release-check so maintainers can see whether commits after the current version tag affect package release contents or only repo-local evidence/docs.

0.34.0

• Changed agentloop install-agent to preserve existing agent instruction files by default and report created/skipped status in human and JSON output.
• Changed default generated agentloop report HTML report paths to preserve same-minute reruns by adding a numeric suffix.
• Changed default generated prepare-pr --write PR description paths to preserve same-minute reruns by adding a numeric suffix.
• Changed default generated verify, ci-summary --write, and release-notes --write artifact paths to preserve same-minute reruns by adding a numeric suffix.
• Changed default create-task path allocation to preserve existing same-day, same-title task contracts by adding a numeric suffix.
• Made human-readable review-context output render dynamic snapshot values on one Markdown line while preserving raw JSON values.
• Made human-readable ship write-confirmation output render written report paths on one Markdown line while preserving raw JSON values.
• Made human-readable schemastore output render catalog names, file matches, and schema URLs on one Markdown line while preserving raw JSON values.
• Made human-readable report output render written report paths, task titles, verification statuses, and changed-file counts on one Markdown line while preserving raw JSON values.
• Made human-readable badge output render written badge paths, sources, statuses, and messages on one Markdown line while preserving raw JSON values.
• Made human-readable verify output render progress commands, written report paths, run paths, and status values on one Markdown line while preserving raw JSON values.
• Made human-readable summarize and handoff output render written summary and run paths on one Markdown line while preserving raw JSON values.
• Made human-readable install-agent output render generated agent-instruction paths on one Markdown line while preserving raw JSON values.
• Made human-readable github import output render dynamic imported metadata values on one Markdown line while preserving raw JSON values.
• Made human-readable upgrade-harness output render dynamic harness audit values on one Markdown line while preserving raw JSON values.
• Made human-readable policy output render dynamic policy and policy-pack values on one Markdown line while preserving raw JSON values.
• Made human-readable create-task output render generated paths and warning command values on one Markdown line while preserving raw JSON values.
• Made human-readable run-ledger output render dynamic run and file-intent values on one Markdown line while preserving raw JSON values.
• Made human-readable task lifecycle output render dynamic task values and diagnostics on one Markdown line while preserving raw JSON values.
• Made human-readable maintainer-check output render dynamic check values and evidence paths on one Markdown line while preserving raw JSON values.
• Made agentloop prepare-pr PR bodies, GitHub comments, and write confirmations render dynamic values on one Markdown line while preserving raw JSON values.
• Made human-readable ship output render dynamic values as single-line inline code or prose while preserving raw JSON values.
• Made human-readable release-proof output render dynamic values as single-line inline code while preserving raw JSON values.
• Made human-readable release-notes output render dynamic values as single-line inline code while preserving raw JSON values and authored changelog prose.
• Made human-readable ci-summary output render dynamic values as single-line inline code while preserving raw JSON values.
• Made human-readable release-check output render dynamic values as single-line inline code while preserving raw JSON values.
• Made human-readable artifacts output render dynamic values as single-line inline code while preserving raw JSON values.
• Made human-readable status and next output render dynamic values as single-line inline code while preserving raw JSON values.
• Made human-readable check-gates output render dynamic values as single-line inline code while preserving raw JSON values.
• Added a Package name doctor check and made human-readable doctor output render dynamic values as single-line inline code while preserving raw JSON values.
• Added shell completion values for artifact types, badge sources, summarize and handoff formats, and the task archive --status done value.
• Added shell completion values for agentloop release-proof --only channels.
• Added agentloop release-proof --only <channel> for targeted npm, GitHub Release, GHCR, or MCP Registry proof checks.
• Changed imported GitHub metadata normalization to bound long titles, states, URLs, authors, labels, branch names, and body excerpts before storage or rendering.
• Fixed local policy-pack reading so symlinked policy files must resolve inside the pack's policies/ directory and the repository.
• Hardened public-doc hygiene checks against unsupported channel claims, fake adoption copy, testimonial language, and premature Pro/SaaS upgrade copy.
• Added agentloop task doctor warnings for open task contracts that still contain AgentLoopKit placeholder text in review-critical sections.
• Added --redact-paths to agentloop verify so verification reports and run-ledger copies can be shared without exposing local absolute roots from command output.
• Added --redact-paths to agentloop summarize and agentloop handoff so reviewer Markdown can be shared without exposing local absolute roots.
• Changed agentloop verify --task-commands to accept verification commands written as Markdown inline-code list items.
• Added a repo-local autonomous dogfood guide for AgentLoopKit, ProjScan, AgentFlight, product-panel review, and simulated research cycles.
• Added AgentFlight health to the project dogfood gate while keeping the script read-only and release-safe.
• Added npm run maintenance:check for the recurring near-term guard covering unit checks, public-doc hygiene, link checks, and strict dogfood.
• Added maintainer guard documentation for release proof, SchemaStore freshness, small policy packs, and read-only GitHub metadata.

0.33.0

• Added agentloop release-proof to check post-release npm, GitHub Release, GHCR, and MCP Registry evidence without publishing, tagging, uploading, or reading tokens.
• Added release-proof documentation and generated harness guidance for post-release evidence collection.
• Changed public-doc hygiene to block fake adoption language, premature Pro/SaaS upgrade copy, and internal planning docs from public-doc scans.
• Added local GitHub metadata summaries to review-context, prepare-pr, and maintainer-check from .agentloop/github/context.json.
• Added organization policy-pack workflow examples for repo-local team review rules.
• Fixed imported GitHub metadata reading so normalized bodyExcerpt fields survive the import-to-review round trip.

0.32.1

• Changed release-status documentation to record verified 0.32.1 npm, GitHub release, GHCR, and MCP Registry proof.
• Added AgentLoop dogfood evidence for the 0.32.1 release gate and archived the completed release task.

0.32.0

• Added agentloop schemastore and agentloop schemastore --json for a ready-to-submit SchemaStore catalog entry without calling GitHub APIs or writing files.
• Added optional local policy-pack config plus agentloop policy packs, agentloop policy pack show, and agentloop policy pack apply for safe bundled and repo-local policy guidance.
• Added bundled agentloop-baseline and maintainer-review policy packs that copy missing policy files without overwriting existing repo policies.
• Added agentloop github import for explicit local issue and pull request JSON import into .agentloop/github/context.json without tokens, API calls, or env-file reads.
• Added docs for SchemaStore support, local GitHub metadata import, Windows package-manager design, and editor-extension validation gates.
• Changed test:unit to include the fast SchemaStore, policy-pack, GitHub metadata, roadmap-channel, and CLI docs drift checks.

0.31.0

• Added agentloop upgrade-harness --details and --suggestions for copyable current-loop guidance in older repos.
• Changed agentloop doctor to recommend agentloop upgrade-harness --details when generated guidance misses current review-readiness topics.
• Added test:unit, test:integration, test:release, and release-flow scripts so maintainers can use a faster dev loop while keeping the full release gate intact.
• Added docs/upgrading-existing-repos.md for safe older-repo upgrades.
• Expanded MCP client setup docs for Claude Code, Cursor, Gemini CLI, OpenCode, and Codex.
• Added a bugfix PR example and expanded the dependency-upgrade example around create-task, verify, ship, and prepare-pr.

0.30.0

• Added agentloop upgrade-harness, a read-only existing-repo upgrade audit with --redact-paths for stale generated guidance around ship, prepare-pr, run ledger, file intent, review-context, and maintainer checks.
• Added a faster test:quick maintainer script for focused local sanity checks while keeping the full Vitest suite as the release gate.
• Added the harness upgrade audit to the project self-check gate so stale generated guidance is caught before release prep.
• Changed getting-started, template migration, MCP, contributor, generated harness, and README guidance to promote the current verify -> ship -> prepare-pr -> maintainer-check review-readiness loop.
• Added dependency audit to the project self-check gate so dependency advisories are caught before release prep.

0.29.0

• Added run ledger entries to agentloop artifacts, including --type run and --latest support, so agents can discover local run evidence from the same inventory command.
• Added agentloop release-check --redact-paths so release-readiness output can be pasted into public issues, PRs, or CI logs without exposing the local Git root.
• Added agentloop maintainer-check --redact-paths so reviewer-readiness output uses the same public-log safety flag as doctor, status, gates, ship, and PR preparation.
• Added agentloop doctor --redact-paths for safe public sharing of setup, Git-root, and risk-file output while keeping default doctor JSON unchanged for scripts.
• Changed pnpm dependency resolution to use patched esbuild@0.28.1 for release tooling and test/build dependencies.
• Changed agentloop ship report and GitHub-comment readiness lists to escape Markdown control characters in warning, blocker, and next-action prose.
• Changed agentloop prepare-pr to escape Markdown control characters in task-derived PR list prose and readiness comment lists.
• Changed dogfood JSON summaries to redact the current workspace root from step arguments, command text, and child-process error messages.
• Changed the dogfood gate to run agentloop review-context --json --redact-paths, matching the public-log safety behavior used by status and gates.
• Added public-docs hygiene coverage that keeps README --redact-paths guidance aligned with supported shareable CLI commands.
• Changed agentloop review-context to accept --redact-paths, matching the shareable-output behavior of status and gate snapshots.
• Changed agentloop next to accept --redact-paths in human and JSON output, matching the shareable-output behavior of status.
• Added public-docs hygiene coverage for stale root FINAL_HANDOFF.md current-release and install guidance, and refreshed the current handoff copy.
• Changed agentloop verify --task-commands to use the active task when --task is omitted, and to exit before running commands when no explicit or active task exists.
• Fixed generated handoff and ship report writes so repeated same-minute runs allocate suffixed Markdown artifact paths instead of overwriting earlier evidence.
• Changed git status collection to request individual untracked files, so new folders appear as file-level evidence in status, ship, handoff, run-ledger, and maintainer outputs.
• Fixed agentloop status and agentloop next next-action copy so covered handoff evidence uses correct grammar.
• Changed agentloop status and agentloop next so active tasks with passing verification and fresh handoff coverage recommend agentloop task done instead of asking for another handoff.
• Changed agentloop maintainer-check to warn when dirty files are not covered by the latest handoff or ship run, instead of accepting any older PR summary as current evidence.
• Fixed agentloop ship so its report and JSON output score the post-ship review-evidence state instead of showing a stale-handoff warning for evidence generated by the same command.
• Fixed agentloop check-gates so agentloop ship remains fresh handoff evidence when its run covers the current dirty files.
• Changed agentloop check-gates to warn when dirty files are not covered by the latest handoff run.
• Added JSON dogfood summaries through npm run dogfood:json and npm run dogfood:strict:json for agents and CI logs.
• Hardened post-verification gate detection so AgentLoop command words inside another shell command do not trigger noisy warnings.
• Broadened post-verification gate detection so create-task and task doctor warn when common AgentLoop review-readiness commands are listed as verification commands.
• Added agentloop task doctor diagnostics for task contracts that put likely post-verification gates under Verification Commands.
• Fixed agentloop check-gates so clean passing evidence reports no required next command instead of asking for another handoff.
• Added a human-readable task done next step that points users to handoff evidence before archival.
• Added the same handoff next-step guidance to successful human-readable bulk task archive output.
• Added a human-readable task archive next step that points maintainers to agentloop handoff --write-run.
• Fixed agentloop handoff so it can preserve archived completed task context from recent run evidence when no active task exists.
• Fixed agentloop check-gates next-action guidance so fresh handoff evidence no longer conflicts with agentloop status.
• Fixed agentloop status so a fresh handoff run does not keep recommending another handoff only because its own run folder is dirty.
• Added a read-only public docs hygiene check to npm run dogfood and npm run dogfood:strict, with a direct npm run check:public-docs script.
• Added create-task warnings when verification commands look like post-verification gates such as dogfood:strict, check-gates --strict, or release-check --strict.

0.28.7

• Added agentloop release-notes --public for concise release-page Markdown that omits local changed-file and AgentLoop evidence inventories.

0.28.6

• Changed agentloop release-check to warn when the latest generated release notes do not mention the local package version.

0.28.5

• Changed agentloop release-check next-action guidance to route maintainers through agentloop npm-status before publishing instead of recommending a direct npm publish command.
• Changed agentloop check-gates --strict so a clean Git tree with current task, verification, and handoff evidence passes instead of failing on the informational No changed files detected message.

0.28.4

• Added agentloop verify --progress for opt-in, bounded per-command progress output during long verification runs while keeping JSON output parseable.
• Added release-smoke coverage that catches stale ROADMAP.md current-release state before packing a release.
• Added built-CLI smoke coverage for agentloop verify --progress.
• Added agentloop runs --latest and agentloop runs --limit <count> for bounded local run-ledger navigation.
• Added built-CLI smoke coverage for bounded agentloop runs output.
• Added agentloop create-task --include-config-commands to copy configured test, lint, typecheck, and build commands into new task contracts without running them during creation.
• Added duplicate-command reporting to verification Markdown and JSON output when exact command strings are selected more than once.
• Changed verification command selection to run exact duplicate command strings once when configured commands and task contract commands overlap.
• Changed Vitest's bounded per-test timeout to 90 seconds so full release gates can run subprocess-heavy CLI integration tests reliably.
• Fixed run ledger summaries so archived task contracts hydrate latest-run task title, status, and path while stored run metadata remains unchanged on disk.
• Fixed agentloop status next-action guidance after archived task evidence so dirty shipped work points back to handoff/review instead of starting a new task.

0.28.3

• Fixed agentloop ship so archived latest-run task contracts still count as task evidence after normal task cleanup.
• Fixed agentloop prepare-pr so it keeps the archived task title, acceptance criteria, risk notes, and rollback notes after cleanup, and reuses fresh ship evidence instead of writing duplicate ship runs.

0.28.2

• Added agentloop verify --task <path> --task-commands --only-task-commands for focused task-contract verification without configured repo commands.
• Changed the README quickstart and verification docs to show explicit task paths when running task contract commands.

0.28.1

• Added npm run dogfood and npm run dogfood:strict as repeatable local self-check gates for AgentLoopKit development.
• Added official AgentLoopKit logo assets under docs/logo/ and refreshed README launch screenshots to use the official mark.
• Changed README, contributor guidance, and agent guidance to document the dogfood gate without adding publish, tag, token, or verification side effects.
• Changed release status and release hygiene docs to keep public package state clear after 0.28.0.
• Archived shipped internal task contracts and added release-smoke coverage that blocks stale planned-release guidance in public-facing harness docs.

0.28.0

• Added agentloop ship, a local review-readiness flow that writes a ship report, records a run ledger entry, and scores task clarity, scope control, verification evidence, evidence freshness, gates, handoff readiness, and risk flags without claiming to measure code quality.
• Added agentloop ship --github-comment to print compact review-readiness Markdown for CI-managed pull request comments without GitHub token handling inside the CLI.
• Added agentloop prepare-pr to generate PR title/body text and GitHub-comment Markdown from local AgentLoopKit evidence without reading GitHub tokens or posting comments.
• Added .agentloop/runs/, agentloop runs, agentloop show-run, and agentloop intent <file> for local run evidence and file-intent lookup.
• Added agentloop verify --write-run, agentloop summarize --write-run, and agentloop handoff --write-run so narrower evidence flows can opt into local run ledger records.
• Added agentloop maintainer-check, a read-only maintainer reviewability check for agent-assisted PRs.
• Added read-only MCP tools for the latest ship report and recent run ledger entries.
• Added read-only MCP agentloop_show_run for fetching one local run ledger entry by id.
• Added read-only MCP agentloop_file_intent for local run-ledger intent lookup by file path.
• Added read-only MCP agentloop_maintainer_check for local maintainer reviewability checks.
• Added read-only MCP agentloop_policy_status for local policy template status.
• Added read-only MCP agentloop_check_gates for local review gate status.
• Added read-only MCP agentloop_artifacts for local artifact inventory metadata.
• Added read-only MCP agentloop_review_context for one local reviewability snapshot that combines status, gates, policies, artifacts, recent runs, and latest ship evidence without artifact bodies.
• Added agentloop review-context and agentloop review-context --json for non-MCP agents that need the same read-only local reviewability snapshot.
• Changed README launch visuals and terminal demo to show the current review-readiness loop with task-aware verification, ship, prepare-pr, review-context, run history, and file intent lookup.
• Changed public JSON and human output for verification, handoff, ship, prepare-pr, and run paths to report repo-relative .agentloop/... paths instead of absolute local filesystem paths.
• Added --redact-paths to agentloop status and agentloop check-gates so public logs can hide the absolute Git root while default JSON remains script-compatible.
• Added --redact-paths to agentloop ship and agentloop prepare-pr so acceptance-layer output can hide the absolute Git root inside embedded gate evidence.
• Changed agentloop status Markdown, JSON, and brief output to include the newest local run ledger entry when .agentloop/runs/ exists.
• Changed agentloop prepare-pr to reuse a matching fresh ship run instead of writing a duplicate run ledger entry.
• Added shipEvidence.source and shipEvidence.runId to agentloop prepare-pr --json output.
• Added agentloop task done as a shortcut for marking the active task, or an explicit task path, as done.
• Changed agentloop prepare-pr PR bodies to group changed files by review area.
• Changed agentloop ship reports/comments and agentloop prepare-pr PR-facing Markdown to render AgentLoop artifact paths repo-relative while keeping JSON path fields script-friendly and repo-relative.
• Changed agentloop create-task to set the newly created contract as the active task and include activeTask in JSON output.
• Changed agentloop status and agentloop next to recommend marking clean, verified review tasks as done before starting unrelated work.
• Changed agentloop policy list/status human output to render policy titles, statuses, and paths with Markdown-safe inline-code delimiters while keeping JSON output and raw policy show content unchanged.
• Changed run ledger reads and writes to reject run-id directories that resolve outside .agentloop/runs/ through symlinks.
• Changed run ledger writes to keep same-minute same-command records by appending collision-safe numeric suffixes instead of overwriting the existing run.
• Changed run ledger reads and writes to return safe display paths for task, artifact, and changed-file metadata, collapsing AgentLoop artifacts to .agentloop/... and outside absolute paths to basenames. MCP run tools and review-context now inherit that behavior from the ledger reader.
• Fixed agentloop prepare-pr so PR bodies include every acceptance-criteria bullet from the task contract instead of only the first line.
• Changed release-note missing-ref fallback copy to render requested Git refs with Markdown-safe inline-code delimiters while preserving raw fallback reason data.
• Changed release notes to render commit subjects with Markdown-safe inline-code delimiters while preserving raw commit subjects in JSON output.
• Changed PR summaries and handoffs to render non-empty diff stats inside Markdown-safe text fences while keeping the no-diff fallback readable.
• Changed PR summaries and handoffs to render the top-level task context title with Markdown-safe inline-code delimiters while keeping verification status text parseable.
• Changed verification report top-level timestamp, repo, git branch, git commit, and working-tree values to render with Markdown-safe inline-code delimiters while keeping overall status parseable.
• Changed verification report task-context path, title, task type, and status values to render with Markdown-safe inline-code delimiters.
• Changed agentloop npm-status Markdown output to render registry error text as a single Markdown-safe inline-code value while preserving exact structured error data in JSON output.
• Changed CLI write-confirmation output for task creation, verification, handoffs, reports, badges, CI summaries, release notes, and agent installation to render generated paths, statuses, sources, messages, and counts with Markdown-safe inline-code delimiters.
• Changed agentloop task human output to render task titles, statuses, paths, lifecycle messages, and task-doctor diagnostics with Markdown-safe inline-code delimiters.
• Changed task-state CLI regression coverage to give aggregate subprocess-loop tests an explicit timeout budget, reducing false failures under full verification load.
• Changed agentloop doctor Markdown output to render check statuses, names, messages, path-like values, overall status, strict mode, and next-action text with Markdown-safe inline-code delimiters.
• Changed agentloop release-check Markdown output to render package labels, git metadata, check details, paths, status values, changed-file counts, and next-action commands with Markdown-safe inline-code delimiters.
• Changed agentloop ci-summary Markdown output to render generated timestamps, evidence titles, evidence paths, gate details, status values, and next-action commands with Markdown-safe inline-code delimiters.
• Changed agentloop check-gates Markdown output to render gate, git, path, status, and next-action command values with Markdown-safe inline-code delimiters.
• Changed agentloop artifacts Markdown output to render task, verification, handoff, HTML report, badge, CI summary, and release-note inventory values with Markdown-safe inline-code delimiters.
• Changed agentloop status and agentloop next Markdown output to render project, git, task, report, command, and working-tree values with Markdown-safe inline-code delimiters.
• Changed release notes to render package, version, range, branch, commit, and AgentLoop evidence values with Markdown-safe inline-code delimiters.
• Changed verification reports and CI summaries to render allowlisted CI metadata values with Markdown-safe inline-code delimiters.
• Refactored Markdown inline-code and code-fence formatting into a shared tested helper used by verification reports, PR summaries, and release notes.
• Changed npm status reports to render package and version labels with Markdown-safe inline-code delimiters when captured registry data contains backticks.
• Changed CLI subprocess regression tests to use a shared timeout budget so full-suite load does not create false timeout failures.
• Changed release notes to render changed-file and working-tree path labels with Markdown-safe inline-code delimiters when paths contain backticks.
• Changed PR summaries and handoffs to render changed-file paths with Markdown-safe inline-code delimiters when paths contain backticks.
• Changed verification report command labels to use Markdown-safe inline-code delimiters when command strings contain backticks.
• Changed verification reports to wrap command output in Markdown fences that are longer than any backtick run in the output, so logs containing ``` cannot break report structure.
• Added built-CLI smoke coverage for missing setup errors and nested working-directory flows across status, create-task, verify, handoff, check-gates, policy, and install-agent.
• Changed packed-release smoke checks to run child commands with an allowlisted environment instead of inheriting token-like variables from the parent shell.
• Changed agentloop npm-status --registry-json to reject .env and .env.* paths before reading captured registry output.
• Changed commands that require AgentLoop setup so missing agentloop.config.json is reported as a clear CONFIG_ERROR with an agentloop init hint instead of a raw filesystem error.
• Changed non-init repo commands to search upward for the nearest agentloop.config.json and run against that initialized root, so status, create-task, verify, handoff, and related commands work from nested source folders without writing nested .agentloop/ directories.
• Added agentloop doctor --strict so CI or team setup gates can treat doctor warnings as failures.
• Added agentloop status --brief for compact task, verification, working-tree, and next-action output.
• Added agentloop verify --timeout-ms <ms> so stuck or watcher-style commands fail with clear timeout evidence instead of hanging indefinitely.
• Changed agentloop init --local-only to ask Git for the real metadata directory before writing info/exclude, so a fake or symlinked .git directory cannot redirect the local-only exclude write.
• Changed review gates, handoffs, release notes, and release checks to ignore stale verification reports that predate the active or newest open task unless the task is already in review or done.
• Changed agentloop release-check to warn when CHANGELOG.md still has pending Unreleased entries before recommending publish steps.
• Changed agentloop npm-status to validate package names before running npm and to clarify that npm may use normal npm configuration during live registry checks.
• Changed agentloop release-notes to reject option-shaped Git refs and pass selected refs after Git's --end-of-options.
• Changed the composite GitHub Action to run through a Node wrapper that validates inputs and uses argument arrays instead of shell interpolation for the AgentLoop command.
• Changed the MCP Registry publish workflow to download a pinned mcp-publisher release and verify its SHA-256 before OIDC authentication.
• Changed agentloop doctor risk-file scanning to use bounded traversal and report when large-repo scanning is truncated.
• Added agentloop release-check, a read-only local release readiness check for package metadata, changelog entries, release scripts, git state, verification evidence, handoff evidence, generated release notes, and npm package-safety basics.
• Added agentloop artifacts --type <type> and agentloop artifacts --latest for filtered local evidence inventories without reading artifact file contents.
• Added actionable doctor next steps in human and JSON output for stale harness metadata, missing verification commands, Git context issues, monorepo warnings, dirty working trees, and risk-file categories.
• Added a CLI docs drift test so the public command surface stays aligned across CLI help, README, CLI reference, and shell completions.
• Changed the composite GitHub Action to support install-mode: npm and install-mode: local, install in the configured working directory, and avoid package-lock rewrites in npm mode.
• Added an end-to-end workflow example that walks through init, doctor, task creation, verification, handoff, gates, artifact inventory, and release readiness.
• Added agentloop artifacts and agentloop artifacts --json, a read-only local evidence inventory for task counts, latest verification report, latest handoff, HTML reports, badges, CI summaries, and release notes.
• Added a cross-platform CLI smoke workflow for Ubuntu, macOS, and Windows that builds the package and exercises the built CLI in a temporary repository.
• Simplified the npm-facing README and moved detailed command behavior into docs/cli-reference.md.
• Improved generated first-run guidance with a risk-aware task contract example, task-linked verification command, and read-only artifact inventory hints.
• Added a VS Code/Open VSX extension design that defers implementation until a thin CLI-wrapper extension has clear demand.
• Added repeatable agentloop create-task --risk and --risk-note flags for non-interactive Risk Notes.
• Changed agentloop check-gates so required root, harness, and policy files only satisfy review gates when their resolved path stays inside the current repo; unsafe symlinked files are treated as missing.
• Changed read-only artifact discovery so task lists, status, gates, handoff fallbacks, HTML reports, badges, CI summaries, release notes, MCP tools, and policy reads ignore configured task, report, handoff, or policy roots that resolve outside the current repo through symlinks.
• Simplified public release and launch docs around the current trusted-publishing flow, and added packed-release smoke coverage that rejects unsupported install-channel claims or maintainer-only release chatter in normal public docs.
• Added packed-release smoke coverage for symlinked init harness targets and symlinked task archive destinations.
• Changed agentloop task archive so the archive destination must resolve inside the configured task directory. With --json, unsafe symlinked archive directories return OUTPUT_PATH_INVALID before moving the task file.
• Changed .agentloop/state.json handling so task active-state reads ignore unsafe symlinked state paths, while task set --json and task clear --json return OUTPUT_PATH_INVALID before writing or removing state outside the current repo.
• Changed agentloop init so generated harness targets must resolve inside the current repo before any init files are written. With --json, unsafe symlinked .agentloop/, AGENTS.md, AGENTLOOP.md, or config targets return OUTPUT_PATH_INVALID.
• Changed agentloop install-agent so .agentloop/agents/*.md and AGENTS.md must resolve inside the current repo before the command reads or writes them. With --json, unsafe symlinked targets now return OUTPUT_PATH_INVALID and no agent files are written.
• Changed generated artifact writes so configured task, report, and handoff roots must resolve inside the current repo even when those directories already exist as symlinks. verify --json, handoff --json, report --json, badge --json, ci-summary --json --write, and release-notes --json --write now return OUTPUT_PATH_INVALID before writing outside the repo.
• Changed agentloop.config.json validation so configured AgentLoopKit paths must be non-empty repo-relative paths, rejecting absolute paths, Windows drive-qualified paths, parent traversal, and null bytes before artifact commands read or write configured locations.
• Changed task artifact path validation to resolve existing symlinked ancestors before writing or reading task files, so agentloop create-task --out <path> and explicit task paths such as agentloop verify --task <path> cannot be redirected outside the configured task directory.
• Changed task lifecycle path validation to resolve existing symlinked ancestors before reading, pinning, updating, or archiving task contracts, so agentloop task show, set, status, and archive cannot operate on files outside the configured task directory through symlinked task paths.
• Changed agentloop check-gates so task-hygiene warnings recommend agentloop task doctor after required task, verification, and handoff evidence exists.
• Added a task-hygiene warning gate to agentloop check-gates so review gates surface agentloop task doctor diagnostics without mutating task files.
• Added CLI regression coverage for invalid explicit artifact output extensions on agentloop report, agentloop badge, agentloop ci-summary --write, and agentloop release-notes --write.
• Changed explicit output paths for agentloop report, agentloop badge, agentloop ci-summary --write, and agentloop release-notes --write so generated artifacts must stay inside their configured AgentLoop artifact directories, with OUTPUT_PATH_INVALID JSON errors for invalid paths.
• Changed agentloop ci-summary --out <path> and agentloop release-notes --out <path> to fail fast unless --write is also passed, with OUT_REQUIRES_WRITE JSON errors for automation.
• Added Baseframe Labs author metadata to the npm package and README.
• Changed agentloop status and agentloop next so unpinned open task contracts appear as latestTask, not activeTask, and the next action asks users to pin the task with agentloop task set <path>.
• Added deferred as a supported task status for parked work that stays visible in task list, status, and next, while staying out of fallback task selection.
• Added deferredTasks to agentloop status --json and agentloop next --json, plus Markdown output that shows parked deferred tasks separately from active work.
• Added agentloop npm-status --agentloopkit so maintainers can check the published AgentLoopKit package from release smoke directories without reading the current folder package name.
• Updated release-smoke helper tests so static analysis can see the helper usage and projscan doctor reports a clean health check.
• Added target directory, detected project, and configured command context to agentloop init and agentloop init --dry-run --json.
• Added non-blocking Git detection to agentloop init human and JSON output.
• Added Git root and target-is-root context to agentloop init human and JSON output.
• Added a human warning when agentloop init targets a Git repository subdirectory.
• Added JSON error output for agentloop init --local-only --json setup failures when the target is not a Git repository.
• Added Git root and target-is-root context to agentloop doctor human and JSON output.
• Added Git root and target-is-root context to agentloop status human and JSON output.
• Added Git root and target-is-root context to agentloop check-gates human and JSON output.
• Added test-generation as a supported create-task --type value and documented the supported task type list.
• Changed create-task --type <value> to fail fast on unsupported types instead of entering the interactive picker.
• Added supported task type values to agentloop create-task --help.
• Added JSON error output for unsupported agentloop create-task --type values when --json is requested.
• Added create-task --type values to generated bash, zsh, fish, and PowerShell completion scripts.
• Added agentloop list-templates --json for machine-readable bundled template discovery.
• Added agentloop version --json for machine-readable version checks.
• Added agentloop install-agent <agent> --json and agentloop install-agent all --json for machine-readable agent setup results.
• Added JSON error output for unsupported agentloop install-agent <agent> values when --json is requested.
• Added JSON error output for unsupported agentloop task status <path> <status> values when --json is requested.
• Added JSON error output for unsupported agentloop badge --source <source> values when --json is requested.
• Added JSON error output for missing agentloop policy show <policy> names when --json is requested.
• Added JSON error output for missing .agentloop/policies/ directories on agentloop policy subcommands when --json is requested.
• Added JSON error output for invalid task paths on agentloop task show, set, status, and archive when --json is requested.
• Added JSON error output for invalid agentloop create-task --out <path> values when --json is requested.
• Added JSON error output for invalid explicit artifact paths on agentloop summarize, handoff, and report when --json is requested.
• Added JSON error output for missing or malformed agentloop npm-status --registry-json files when --json is requested.
• Added JSON error output for invalid agentloop npm-status --timeout-ms values when --json is requested.
• Added JSON error output for invalid agentloop verify --task paths when --json is requested.
• Added JSON error output for invalid config on agentloop status, next, and check-gates when --json is requested.
• Added JSON error output for invalid config on agentloop verify, summarize, handoff, report, badge, ci-summary, and release-notes when --json is requested.
• Added JSON error output for invalid config on agentloop create-task, task, and policy commands when --json is requested.
• Changed malformed agentloop.config.json parsing failures to return CONFIG_ERROR, matching schema-invalid config behavior.
• Added unsupported-format validation for agentloop summarize and agentloop handoff, with JSON errors when --json is requested.
• Added --verification <path> as an alias for --report <path> on agentloop summarize and agentloop handoff.
• Added --verification <path> as an alias for --report <path> on agentloop report.
• Changed the composite GitHub Action default agentloopkit-version to latest; workflows can still pin <version> for reproducibility.
• Added GitHub Action guidance warning maintainers not to pass untrusted pull request or user input to the composite action command.
• Changed the composite GitHub Action install step to pass agentloopkit-version through an environment variable instead of direct shell interpolation.
• Documented the maintainer rule to accumulate active development for the planned 0.28.0 batch instead of cutting a release for every small improvement.
• Changed npm run smoke:release so the packed README may stay unpinned while stale exact version pins are still rejected.
• Added agentloop verify --task-commands to explicitly run commands listed in a task contract's Verification Commands section.
• Added a verification report note when --task-commands is requested but no runnable task commands are found.
• Added taskCommands.requested and taskCommands.foundCount to agentloop verify --json.
• Added taskCommands.commands to agentloop verify --task-commands --json so automation can inspect selected task-defined commands without parsing Markdown.
• Added Vitest coverage that rejects hardcoded AgentLoopKit version pins in normal public docs and examples while allowing release-history evidence files.
• Replaced stale current-version pins in user-facing README, MCP, distribution, CI docs, and examples with @latest or <version> placeholders.
• Added regression coverage for missing Git executable behavior in local Git helpers and agentloop init.
• Removed internal product-panel notes from the public distribution-channel guide, archived completed trusted-publishing evidence, and cleared stale proposed task contracts from the active task folder.
• Archived the remaining completed and legacy AgentLoopKit task contracts so agentloop task doctor reports a clean active task folder.

0.27.0

Minor release for task-folder hygiene diagnostics:

• Added agentloop task doctor, a read-only task folder hygiene check for missing, legacy, unsupported, and terminal task statuses.
• Added task doctor to shell completions, README usage, task docs, generated harness guidance, and bundled agent instructions.

0.26.5

Patch release for terminal task fallback cleanup:

• Fixed fallback task selection so status, next, gates, handoffs, HTML reports, CI summaries, and release notes ignore unpinned terminal tasks marked done, completed, or verified.
• Kept explicitly pinned done tasks visible so status and next recommend agentloop task archive <path>.

0.26.4

Patch release for SchemaStore editor discovery:

• Registered agentloop.config.json in SchemaStore for editor auto-discovery: https://github.com/SchemaStore/schemastore/pull/5783.
• Added the SchemaStore note to the npm-facing README and release-channel docs.

0.26.3

Patch release for status cleanup guidance:

• Updated agentloop status and agentloop next to recommend agentloop task archive <path> when the pinned active task is already done.

0.26.2

Patch release for release-channel cleanup and agent guidance:

• Added a specialist agent roster to generated AGENTS.md and this repository's own AGENTS.md, with roles for product, CLI, templates, verification, security, release, docs, compatibility, MCP, and repo stewardship.
• Documented the generated roster in the README and added init coverage so future template changes keep it in place.
• Removed the temporary Homebrew tap/formula channel from release docs and distribution tests. npm/npx remains the primary install path.

0.26.1

Patch release for MCP Registry metadata validation:

• Shortened the MCP Registry server description to satisfy the registry's 100-character limit
• Updated 0.26.x docs and install examples to point at agentloopkit@0.26.1
• Kept agentloop mcp-server read-only with no command execution, file mutation, telemetry, external API calls, or .env content reads

0.26.0

Minor release for MCP and distribution channels:

• Added agentloop mcp-server, a read-only MCP stdio server for local AgentLoopKit status, tasks, policies, verification reports, and handoffs
• Added MCP Registry metadata through server.json and package.json mcpName
• Added a post-npm MCP Registry publish workflow that uses GitHub OIDC and verifies the npm package version before publishing metadata
• Added a root GitHub Action wrapper for running AgentLoopKit commands in CI
• Added a Dockerfile and GHCR release workflow for a minimal agentloop container image
• Added a Homebrew formula template under packaging/homebrew/
• Added docs for MCP usage, distribution channels, GitHub Action usage, Docker/GHCR, and Homebrew publishing
• Added Vitest coverage for MCP tool behavior, MCP stdio startup, and distribution artifacts

0.25.0

Minor release for local-only AgentLoopKit harnesses:

• Added agentloop init --local-only for developers who want AgentLoopKit guidance in a repo without committing generated harness files
• Local-only mode writes a marked block to this clone's .git/info/exclude for .agentloop/, AGENTS.md, AGENTLOOP.md, and agentloop.config.json
• Generated AGENTS.md and AGENTLOOP.md include a local-only notice so agents know not to commit the local harness
• Added dry-run, idempotence, non-Git folder, and packed-package smoke coverage for local-only initialization
• Documented local-only setup in README and getting-started docs

0.24.5

Patch release for release safety and README demo polish:

• Added npm run smoke:release for local packed-package release checks
• The release smoke script builds, packs, runs the packed binary, checks safety guards, and verifies packaged README pins without publishing
• Refreshed the README VHS terminal demo around the first-run loop, verification evidence, handoff, gates, local report, and badge output

0.24.4

Patch release for npm README pin freshness:

• Updated README pinned-version examples from 0.24.2 to the current package line
• Kept this release docs-only with no CLI behavior changes

0.24.3

Patch release for task-path safety:

• Restricted agentloop verify --task to Markdown task contracts inside the configured task directory
• Restricted agentloop create-task --out to Markdown files inside the configured task directory
• Refused agentloop init --dry-run in the user's home directory unless --force is passed, preventing accidental home-directory project detection
• Added regression coverage for task path reads, task output writes, and home-directory dry-run safety

0.24.2

Patch release for safer first-run initialization:

• Bounded fallback project detection so agentloop init --dry-run returns quickly in large metadata-free directories such as a home folder
• Added a home-directory guard so agentloop init refuses to write AgentLoopKit files into the user's home directory unless --force is passed
• Added agentloop init --force for the rare case where a user intentionally wants to initialize the current home directory
• Added regression coverage for bounded fallback detection and home-directory init safety
• Clarified README and getting-started docs to run init from the target repository root

0.24.1

Patch release for first-run project detection:

• Fixed agentloop init crashing when project detection encounters unreadable directories such as macOS .Trash
• Changed recursive project file discovery to skip directories it cannot read instead of aborting the command
• Added regression coverage for unreadable directories during project detection

0.24.0

npm-status release. This is the normal next minor release after v0.23.0; npm later caught up to the current release line after maintainer authentication and trusted publishing were configured:

• Added agentloop npm-status for read-only npm registry catch-up checks
• Added agentloop npm-status --json for release scripts and handoff evidence
• Added agentloop npm-status --expect-current for post-publish smoke checks that fail until npm latest matches local package metadata
• Added agentloop npm-status --registry-json <path> so CI and release handoffs can replay captured npm view output without a network call
• Compared local package.json metadata with npm view <package> version versions --json output
• Documented npm-status usage across README, publishing docs, release-status docs, release notes, generated harness guidance, and AgentLoopKit's own repo harness
• Kept the command read-only: no package publishing, tag creation, GitHub release creation, token reads, .env reads, uploads, or package metadata mutation

0.23.0

PowerShell completion release candidate. npm still serves 0.1.1 until account authentication or trusted publishing is repaired. This is the normal next minor release after v0.22.0; after npm catches up to the current release line, future releases should use ordinary sequential semver:

• Added agentloop completion powershell to print a static PowerShell Register-ArgumentCompleter script
• Added agentloop completion pwsh as an alias for PowerShell users
• Included top-level commands, task subcommands, policy subcommands, task statuses, agent names, and shell names in PowerShell completion output
• Kept completion generation stdout-only with no shell profile mutation, telemetry, network calls, or runtime dependency
• Updated README and getting-started docs so shell-completion guidance includes PowerShell

0.22.0

Task-linked verification and README evidence release candidate. npm still serves 0.1.1 until account authentication or trusted publishing is repaired. This is the normal next minor release after v0.21.0; after npm catches up to the current release line, future releases should use ordinary sequential semver:

• Added a Failure Summary section to failed verification reports with failed command, exit code, and final useful output lines
• Kept full command output excerpts in verification reports; the new summary does not diagnose root cause or parse tool-specific logs
• Added task context to verification reports generated with agentloop verify --task <path>
• Guarded agentloop verify --task so .env-style paths are reported as unavailable instead of read as task contracts
• Refreshed README Playwright screenshots and VHS terminal demo around task-linked verification evidence

0.21.0

Next-action and release-safety release candidate. npm still serves 0.1.1 until account authentication or trusted publishing is repaired. This is the normal next minor release after v0.20.0; after npm catches up to the current release line, future releases should use ordinary sequential semver:

• Added agentloop next for a read-only next-action view backed by the existing status engine
• Added agentloop next --json for agents and scripts that need the next command, reason, active task, latest verification report, and dirty-state context
• Updated status and next-action logic so a verification report older than an active in-progress task no longer counts as current task evidence
• Added a prepublish metadata guard that fails while CHANGELOG.md has unreleased entries
• Added agentloop release-notes --release-version <version> so release drafts can override package metadata without colliding with the CLI's global --version flag
• Documented that next does not run verification commands, mutate task state, call an LLM, make network requests, or read .env contents

0.20.0

Release-note handoff release candidate. npm still serves 0.1.1 until account authentication or trusted publishing is repaired. Because GitHub releases already exist from v0.2.0 through v0.19.0, npm should catch up to this current release line once and then return to normal sequential semver:

• Added agentloop release-notes for deterministic local release-note drafts
• Added agentloop release-notes --json for scripts and CI logs
• Added agentloop release-notes --write to write .agentloop/handoffs/YYYY-MM-DD-HH-mm-release-notes.md
• Included package metadata, git range, changelog section, commits, changed files, working tree status, active task, latest verification report, and latest CI summary when available
• Kept release-note generation local and read-only unless --write is passed: no tag creation, package publishing, provider API calls, token reads, uploads, or changelog rewrites
• Documented the npm catch-up policy so the version jump from 0.1.1 to the current GitHub line is a one-time recovery path, not the ongoing release strategy
• Recorded the release-triggered v0.20.0 Publish workflow failure at npm authorization

0.19.0

CI summary and contributor-onboarding GitHub release. npm publish is pending account authentication or trusted-publishing repair. npm may jump from 0.1.1 to this current release because the intermediate versions were GitHub/source release candidates while npm publishing was blocked:

• Added docs/contributor-playbook.md with copyable good-first issue examples and contributor verification expectations
• Linked contributor playbook guidance from README and CONTRIBUTING.md
• Added agentloop ci-summary for local CI provenance and AgentLoop evidence summaries
• Added agentloop ci-summary --json for scripts and CI logs
• Added agentloop ci-summary --write to write .agentloop/reports/YYYY-MM-DD-HH-mm-ci-summary.md
• Kept CI summaries read-only unless --write is passed: no provider API calls, secret reads, telemetry, uploads, or verification command execution
• Updated verification-report lookup so status, summaries, gates, reports, and badges ignore newer CI summary artifacts
• Recorded the release-triggered v0.19.0 Publish workflow failure at npm authorization

0.18.1

Policy customization guidance patch release candidate. GitHub release v0.18.1 is public; npm publish is pending account authentication or trusted-publishing repair:

• Added policy customization workflow docs and generated harness guidance for policy status results
• Added status-specific maintainer actions for current, modified, missing, and extra policy files
• Clarified that local policy files are repo guidance and bundled templates are comparison material
• Updated generated AGENTS, AGENTLOOP, workspace README, harness commands, and review checklist guidance

0.18.0

Policy template status release candidate. GitHub release v0.18.0 is public; npm publish is pending account authentication or trusted-publishing repair:

• Added agentloop policy status for read-only local policy template comparison
• Added agentloop policy status --json with deterministic current, modified, missing, and extra entries
• Added shell completion coverage for the new policy subcommand
• Documented policy status as template drift visibility, not compliance enforcement
• Recorded the release-triggered v0.18.0 Publish workflow failure at npm authorization
• Recorded the local exact-tarball 0.18.0 npm publish attempt that failed with authorization E404

0.17.0

Policy inspection and template-provenance release candidate. npm publish is pending browser/OTP authentication or trusted-publishing repair:

• Added .agentloop/manifest.json during init to record generated template provenance
• Added agentloop doctor template manifest checks for current, missing, stale, invalid, and newer generated harness metadata
• Added manual template migration guidance without changing config validation or overwriting edited harness files
• Added agentloop policy list and agentloop policy show <policy> for read-only local safety policy inspection
• Added JSON output for policy listing and policy reads

0.16.0

Local evidence badge and reviewer-artifact release. This is the npm catch-up release after GitHub-only release candidates from 0.2.0 through 0.15.1:

• Added deterministic PR summary change-area classification for source, tests, docs, CI, config, AgentLoop, risk-sensitive, and uncategorized paths
• Added path-based review-focus hints to PR summaries without LLM calls or file-content inspection
• Added agentloop report for local static HTML evidence reports built from task, verification, handoff, git, and deterministic summary artifacts
• Kept agentloop report --json compact by returning the output path, metadata, and source paths without embedding the full HTML body
• Made agentloop report ignore bundled handoff templates when looking for the latest generated handoff
• Added agentloop badge for local SVG evidence badges built from verification or gate status

0.15.1

Trust-polish GitHub patch release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added category-level agentloop doctor risk-file warnings with capped path examples
• Kept risk-file findings warning-only and path-only
• Reduced noisy risk-file false positives from Markdown docs and templates
• Replaced the unproven custom config schema URL with the GitHub raw schema URL
• Clarified that CLI config validation is local and does not fetch the schema URL at runtime
• Updated README, configuration docs, product-panel records, and release handoff notes

0.15.0

CI-context GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added allowlisted GitHub Actions context to agentloop verify reports
• Added a generic CI=true fallback for verification reports outside GitHub Actions
• Kept local verification reports quiet when CI is not detected
• Added tests for GitHub Actions metadata, generic CI metadata, local omission, and CI-env test isolation
• Updated README, verification docs, GitHub Actions docs, examples, generated harness guidance, product-panel records, and release handoff notes

0.14.0

check-gates strict-mode GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop check-gates --strict so CI can treat warning gates as failures
• Kept default agentloop check-gates warning behavior unchanged for local use
• Added strict to check-gates --json output
• Added strict-mode text to generated gate-check reports
• Added Vitest coverage for default warning behavior and strict warning failure
• Updated README, getting-started docs, gate-check docs, generated harness guidance, and README visual sources

0.13.0

check-gates GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop check-gates to check review-evidence readiness without running tests
• Added agentloop check-gates --json for deterministic agent and CI usage
• Checks task contract, verification report, handoff summary, repo harness, safety policies, and git context
• Added Vitest coverage for ready and missing-evidence paths
• Updated README, getting-started docs, gate-check docs, generated harness guidance, shell completions, agent templates, and README visual assets

0.12.0

create-task JSON output GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop create-task --json for machine-readable task creation
• JSON output includes the created task path and Markdown content
• Default human-readable create-task output remains unchanged
• Added Vitest coverage for JSON task creation output
• Updated README, getting-started docs, task-contract docs, generated task guidance, and README visual assets for the current task lifecycle

0.11.0

Task archive GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop task archive <path> to move one named task contract into .agentloop/tasks/archive/
• Archive refuses to overwrite an existing archived task file
• Archiving the active task clears the active task pointer
• Normal agentloop task list output excludes archived task contracts
• Updated README, getting-started docs, task docs, generated harness guidance, shell completions, and README visual assets for the archive workflow

0.10.0

Shell completions GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop completion <bash|zsh|fish> to print shell completion scripts
• Added static bash, zsh, and fish completion output with no shell profile mutation
• Completion scripts cover top-level commands, task subcommands, task status values, install-agent names, and supported shell names
• Added Vitest coverage for completion rendering and unsupported shell errors
• Updated README, getting-started docs, and README visual assets for the shell-completion workflow

0.9.0

Task status transitions GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop task status <path> <status> to update a task contract's - Status: line
• Added JSON output for status updates
• Added a fixed status set: proposed, in-progress, blocked, review, and done
• Updated generated AGENTS, AGENTLOOP, harness, task, and agent templates with task-status guidance
• Refreshed README screenshots and VHS terminal demo to show the task lifecycle command

0.8.0

Launch-quality monorepo guidance and docs trust GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added generated monorepo verification guidance to .agentloop/harness/commands.md, .agentloop/README.md, and .agentloop/tasks/README.md
• Updated docs to explain root checks versus package-level checks without claiming workspace orchestration support
• Updated the doctor monorepo warning with package-specific verification command examples
• Added a local Markdown link checker with tests, pnpm check:links, and CI coverage

0.7.0

Monorepo doctor awareness GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added monorepo marker detection to agentloop doctor for package workspaces, pnpm-workspace.yaml, Turbo, Nx, Lerna, and Rush config files
• doctor now reports detected workspace markers as warnings without changing project type or failing setup

0.6.0

Task reading GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop task show <path> and agentloop task show <path> --json to read a selected task contract without changing active state
• Updated docs and generated agent templates to use the task list, task show, task set sequence

0.5.0

Task discovery GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop task list and agentloop task list --json to inspect task contracts before pinning one
• Updated agent and harness templates to guide agents toward task discovery before active-task selection

0.4.0

Active task lifecycle GitHub release candidate. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop task set <path> to pin the active task contract
• Added agentloop task current and agentloop task clear
• Added .agentloop/state.json as a transparent repo-local active task pointer
• Updated agentloop status and deterministic handoffs to prefer the pinned active task before falling back to newest task file

0.3.0

Handoff command GitHub release. npm publish is pending trusted-publishing or local-auth repair:

• Added agentloop handoff as the clearer command for writing reviewer handoffs
• Kept agentloop summarize read-only unless --write is passed
• Updated agentloop status to suggest agentloop handoff when task and verification evidence exist
• Fixed repeated non-interactive create-task flags so all values are preserved
• Added create-task --likely-file and --forbidden-file for non-interactive task scoping
• Added create-task aliases for full task-contract fields: --problem-statement, --desired-outcome, --assumption, --verification, and --rollback
• Updated latest task/report detection to prefer newest modified Markdown files instead of filename order
• Updated first-run docs, generated workspace README guidance, and PR summary docs

0.2.1

Review-safety GitHub release. npm publish is pending trusted-publishing or local-auth repair:

• Verification reports now preserve the beginning and ending output when a command log is truncated
• Truncated command output includes a clear marker with the original output length
• agentloop status now points back to agentloop verify when the latest verification report failed
• Added release recovery docs for the v0.2.0 GitHub/npm split
• Added GitHub contributor labels and good-first-issue guidance

0.2.0

Status command:

• Added agentloop status for active task, latest verification report, dirty files, configured commands, and next action
• Added agentloop status --json for machine-readable automation
• Fixed agentloop version and agentloop --version to read from package.json
• Added Vitest coverage for status and version behavior

0.1.1

README and release polish:

• Added README hero and verification screenshots rendered with Playwright
• Added a VHS terminal demo for the published npm CLI
• Updated README publishing status now that agentloopkit is live on npm
• Added a publish workflow guard that skips npm publish when the package version already exists
• Updated launch checklist and final handoff status

0.1.0

Initial MVP:

• TypeScript Node CLI
• init, doctor, create-task, verify, summarize, install-agent, list-templates, and version
• AgentLoopKit templates for loops, gates, policies, handoffs, agents, tasks, and harness files
• Deterministic verification reports and PR summaries
• Vitest coverage for core behavior
• GitHub Actions CI

Follow-up launch polish:

• Added agentloop init --dry-run
• Added schema/agentloop.config.schema.json
• Included the schema in npm package contents
• Added GitHub Actions workflow for npm trusted publishing
• Added generated .agentloop/README.md
• Added agentloop install-agent all
• Added internal product panel, personas, simulated review cycles, backlog, dogfood log, and final handoff
• Added launch checklist