What shipped, version by version.

[0.8.0] - 2026-06-21

Added

• Verification runs now store source-free proof snapshots of changed files so Review Intelligence can distinguish current proof from stale proof after a file changes.
• Status, Markdown reports, HTML replays, resume prompts, snapshots, and handoffs now surface proof freshness as current, stale, covered, missing, failed, not required, or unknown.

Changed

• Review focus rows now include a compact proof-status line across local review surfaces, making stale proof visible without opening session JSON.
• Stale proof now creates an actionable proof gap and keeps readiness at Needs verification until the relevant proof command is rerun.

Fixed

• Verification proof capture now avoids slow git-status probing in non-git sessions with no known changed files.

[0.7.1] - 2026-06-21

Fixed

• Refreshed the README terminal hero GIF so it shows the current handoff-first review flow instead of the older replay-first flow.
• Updated the README hero caption and sample output to include handoff and history accurately.

Changed

• Tightened README workflow copy around the local handoff packet, recent local session history, and current failure-excerpt surfaces.

[0.7.0] - 2026-06-21

Added

• Added agentflight handoff, a local-only review handoff command that generates the report, replay, and resume artifacts and summarizes readiness, proof gaps, review focus, and failed verification excerpts.
• Added agentflight history, a read-only local command for listing recent sessions, proof counts, the current-session marker, and existing report/replay artifact paths without search indexing, export, sync, or session switching.
• Added local agentflight history --task <text> and agentflight history --state ready|blocked|needs_verification|unknown|current filters so engineers can find relevant local sessions without scanning long histories or adding any index, sync, export, or session switching behavior.
• Added a compact Review Path section to local HTML replay artifacts so long sessions lead reviewers through proof gaps, unresolved failed runs, review focus, and verification evidence without adding scripts, exports, or hosted behavior.
• Documented the agentflight history --limit 1 latest-action workflow for reopening local handoff/report/replay/resume artifacts.
• Added session-specific handoff artifacts under .agentflight/reports/ so agentflight history can point to stable handoff packets from prior sessions.
• Added session-specific resume artifacts under .agentflight/reports/ so agentflight history can point to stable continuation prompts from prior sessions.
• Added post-v0.6.0 user-research findings and a v0.6.0 website update prompt focused on the local handoff workflow.
• Added a post-v0.6.0 product direction note that keeps local handoff, first-run workspace hygiene, replay ergonomics, proof guidance, and explainable ranking as the priority order.

Fixed

• Ready agentflight status output now points at an existing local handoff, replay, or report artifact when one already exists, instead of repeating handoff-generation guidance.
• Ready agentflight resume prompts now point at an existing local handoff, replay, or report artifact when one already exists, instead of repeating handoff-generation guidance.
• Review Intelligence now ignores unfinished AgentFlight readout/artifact commands such as agentflight replay and node dist/cli.js replay when computing incomplete proof gaps, so readiness guidance keeps pointing at meaningful verification commands.
• Incomplete verification guidance now says a command may still be running and suggests waiting before rerunning, avoiding false lost-evidence alarms when status is checked during parallel verification.
• Ready agentflight handoff terminal guidance now tells reviewers to share the local handoff packet first, with report/replay as supporting detail artifacts.
• Current start-only agentflight history sessions now guide users to run verification before handoff when no proof exists yet, while keeping handoff-only guidance once verification has been recorded.
• Public positioning regression coverage now guards current public/runtime surfaces against stale assistant-style positioning.
• Clean-worktree status now tells users to open handoff/report/replay or JSON output for tucked verification details, matching the handoff-first ready review path.
• Ready-session Open first guidance now prefers the generated handoff packet in handoff, clean status/resume, and history surfaces, while blocked sessions still point to the report/fix path.
• Full Markdown proof reports now show changed files, risk, verification, review focus, proof gaps, readiness, recommendation, and next action before the timeline, keeping long sessions faster to review.
• agentflight history now compacts non-current start-only sessions that have no proof or review artifacts, keeping recent local artifacts easier to scan.
• Review Intelligence now keeps generated .agentflight/.gitignore helper files below real first-run review targets while leaving them visible.
• agentflight resume now includes unresolved and historical failed-run context below the verification count, matching status/report trust cues.
• Clean-worktree agentflight resume prompts now include the latest local Open first: artifact path when current report/replay/handoff evidence exists.
• Clean-worktree agentflight status now shows the latest local Open first: artifact path when current report/replay/handoff evidence exists.
• agentflight history now shows the nearest previous local artifact when the current latest session has none yet, while keeping the current-session handoff next action.
• First-run AgentFlight generated-file lists and local-file guidance now use shared output helpers, keeping init and start --yes copy consistent.
• agentflight start --yes now explains the AgentFlight files it generated during safe auto-init, including project config and local runtime evidence.
• agentflight init and agentflight start now use concise ProjScan version checks instead of extra help probing, while agentflight doctor keeps deeper diagnostics.
• agentflight doctor now renders multiple detected proof-command suggestions as an indented list instead of one long semicolon-separated line.
• agentflight history now prefers useful review-ready or blocked artifact metadata over later clean-worktree artifact metadata, while preserving clean-only session history.
• Clean-worktree agentflight handoff now preserves existing session review artifacts instead of overwriting report/replay/resume evidence with an empty post-commit clean-state artifact.
• Clean-worktree agentflight handoff now exits successfully instead of treating an informational post-commit handoff as a command failure.
• agentflight doctor now suggests concrete detected agentflight verify -- ... commands when config verification commands are empty.
• agentflight verify now suggests detected package proof commands when no explicit command is provided and config commands are empty.
• agentflight doctor now warns when package proof scripts exist but .agentflight/config.json has no configured verification commands.
• agentflight history now lists capped repo-relative malformed session paths instead of only reporting a skipped-file count.
• Concurrent agentflight verify runs now reserve distinct stdout/stderr evidence paths and merge verification updates without dropping either run.
• agentflight history --limit now rejects non-integer, zero, and negative values with a clear local error instead of silently falling back or returning an empty history.
• Review Intelligence now treats an earlier failed verification as resolved when the same stored command later passes, so TDD red/green and format-fix loops do not leave handoffs permanently blocked.
• Status, report, replay, and handoff now distinguish unresolved failed verification from historical failed runs that later passed.
• Resume prompts now use the same unresolved-versus-resolved verification count wording as the other review surfaces.
• Clean-worktree status now points users to agentflight history --limit 1 to reopen the latest local artifacts before starting the next session.
• History now shows unresolved-versus-resolved failed verification counts for prior sessions.
• History now includes stable handoff artifact paths alongside report and replay paths when those artifacts exist.
• History now includes stable resume artifact paths when those artifacts exist.
• History now suggests which existing local artifact to open first for each session.
• History now surfaces the newest session's open-first action before the full session list, reducing scan work in long local histories.
• History now tells users to run agentflight handoff when the latest local session is current but no handoff/report/replay artifact exists yet.
• History now shows the latest session's recorded readiness in the top-level Latest action: block.
• History now says Open first: none yet when no handoff, report, or replay artifact exists.
• HTML replay now reserves urgent failed-run navigation for unresolved failed verification while keeping historical failed runs visible in the ledger.
• Ready handoffs no longer inline historical failed verification excerpts once no unresolved failed-verification proof gap remains; those excerpts stay in report/replay evidence.
• agentflight start now treats AgentLoopKit's Active task: none pinned. status as no active task instead of falsely reporting task reuse.
• AgentLoopKit task-link diagnostics now use generic link-check wording instead of stale automatic task-creation copy.
• agentflight init now reports ProjScan and AgentLoopKit CLI availability using the same concise tool formatter as start/report surfaces instead of relying on repo marker files.
• agentflight status now reports a clean worktree explicitly instead of calling zero changed files Unknown after a completed session has been committed.
• agentflight status now compacts very long terminal verification run lists while keeping full verification runs in JSON, report/replay, and local evidence files.
• Clean-worktree agentflight status now tucks individual verification run details when there are no unresolved failed runs, while keeping counts and JSON evidence complete.
• Review Intelligence now treats incomplete verification attempts as blocking before clean-worktree readiness, so live status cannot call a session clean while verification is still in progress.
• agentflight init now lists created and skipped files by repo-relative path instead of only showing counts.
• agentflight doctor now warns when .projscan-memory/memory.json exists without a matching project filter and reports OK once the repo filters it through changedFileFilters.ignore.
• agentflight history now labels stored review metadata as Recorded readiness: so it is not confused with live worktree readiness.
• agentflight doctor no longer prints the absolute repository root in the successful repository-root check.
• HTML replay now labels resolved failed verification rows as historical when no unresolved failed runs remain.
• agentflight doctor now treats a missing current session as OK first-run guidance instead of warning when the rest of the local setup is healthy.
• Clean-worktree status now reports Risk: none instead of Risk: unknown while preserving unknown for legacy or genuinely missing metadata.
• Clean-worktree risk reasons now use current-state wording instead of saying no changed files were detected "yet."
• Parallel report, replay, and resume commands now preserve each artifact event in session history instead of letting the last stale session write win.
• Review Intelligence no longer lets ProjScan risk hints boost generated .projscan-memory/memory.json above real first-run review targets, while the file remains visible with the existing changedFileFilters.ignore guidance.
• agentflight history now includes the selected local artifact path directly on the Open first: line, reducing lookup work in long session lists.
• agentflight handoff now includes the selected report or replay path directly on the Open first: line while preserving the full artifact list.

Changed

• Clean-worktree agentflight resume constraints now tell agents to start a new AgentFlight session before unrelated work instead of applying active-task constraints to a completed clean session.
• agentflight start now prefers configured no-arg agentflight verify guidance when .agentflight/config.json already has verification commands, while keeping detected package-script fallback guidance for empty configs.
• Current product copy now uses coding agent sessions and agentic engineering language instead of assistant-style positioning.
• Idempotent agentflight init now shows a concrete detected proof command when existing config verification commands are empty.
• First-run agentflight init now points seeded configs at no-arg agentflight verify in the primary workflow.
• agentflight init now points first-run users through the handoff golden path: start a session, capture verification, then generate a local handoff, with status and doctor listed as supporting checks.
• agentflight init now uses the first detected verification command in its primary workflow guidance, falling back to <proof command> when no proof script is detected.
• Newly generated .agentflight/config.json files now seed detected verification commands from package scripts while leaving profiles empty and preserving existing configs.
• README and verification docs now describe the handoff-first first-run workflow and init-seeded verification commands.
• Kept ready-review report, replay, resume, examples, and demo copy aligned with the agentflight handoff golden path while keeping report/replay/resume as supporting local artifacts.
• Changed-file review surfaces now fail with an actionable git-status error instead of treating git-status failures as an empty diff.
• Shortened the optional ProjScan baseline budget during agentflight start so busy local ProjScan work cannot stall session startup for too long.
• agentflight start now reuses an active AgentLoopKit task instead of creating a duplicate AgentFlight placeholder task.
• agentflight start now shows concise ProjScan and AgentLoopKit warning summaries when optional tooling is available but degraded.
• agentflight start now inspects ProjScan availability without running the heavier optional projscan start baseline on the startup path.
• agentflight start now uses lightweight AgentLoopKit availability inspection on the startup path while preserving task reuse/linking.
• agentflight start now reuses AgentLoopKit's local active-task state file directly instead of parsing agentloopkit status output.
• agentflight start now links existing AgentLoopKit task state without creating new AgentLoopKit task contracts automatically.
• Start output and Markdown tooling rows now show whether AgentLoopKit has an active task linked when that local state is known.
• agentflight handoff now treats missing required proof as not ready to share: it exits non-zero, uses Fix before sharing, and points users to the report first.
• Start, report, replay, and handoff terminal output now display local .agentflight/... artifact paths relative to the repo instead of absolute user-directory paths.
• Handoff verification details now distinguish zero verification runs from passing runs that simply have no failed excerpts.
• Review Intelligence suggested proof now follows each proof gap's preferred proof-kind order, so source gaps prefer npm test when available and dependency gaps prefer build/install-style proof before typecheck.
• Review Intelligence proof-gap rules are now centralized in one ordered table to keep future proof guidance changes easier to review.
• HTML replay verification ledgers now display long run commands compactly while keeping the full command available in the title text.
• Status and Markdown report verification evidence rows now use compact display labels for long run commands while preserving stored command evidence.
• agentflight init now writes .agentflight/.gitignore for runtime evidence directories instead of seeding new runtime .gitkeep files, reducing first-run Git noise while keeping .agentflight/config.json and the local AgentFlight ignore file visible as project config.
• AgentFlight session JSON writes now use same-directory temp files and atomic rename so concurrent report, replay, resume, and handoff commands do not read partially written session state.
• Review Intelligence now describes .projscan-memory/memory.json as generated tool state instead of arbitrary unknown code while keeping the file visible.
• Report and replay generation now persist a compact local readiness summary in session events, letting agentflight history show the latest recorded readiness without recalculating old sessions.

AgentFlight v0.6.0 - 2026-06-19

Local review ergonomics and automation surfaces for heavier real-world dogfood.

Added

• Replay navigation for long HTML evidence ledgers, including jump links, section anchors, failed-run anchors, and a shortcut to the first failed verification run.
• Local JSON status output with agentflight status --format json for scripts that need structured session, risk, verification, and Review Intelligence state.
• Named verification profiles with agentflight verify --profile <name> for local command groups stored in .agentflight/config.json.
• Markdown report modes:
  • agentflight report --mode compact for shorter local review summaries.
  • agentflight report --mode pr-comment for a local PR-comment draft that is never posted automatically.
• Optional typed ProjScan review hints for deterministic Review Intelligence ranking enrichment without making Review Intelligence call ProjScan directly.
• First-run guidance that explains which .agentflight/ paths are runtime evidence and which files are project config.

Changed

• Clarified first-run workspace hygiene docs: .projscan-memory/** can be added to changedFileFilters.ignore when ProjScan memory is generated evidence rather than a review target.

• Lowered generated ProjScan memory priority in Review Intelligence so .projscan-memory/memory.json remains visible but no longer outranks real first-run review targets such as .agentflight/config.json or docs changes.

• Classified first-party TypeScript/JavaScript source files under src/ as source changes so review focus gives clearer guidance and proof gaps than the previous unknown-file fallback.

• Aligned ready-review next actions with the handoff golden path: status, report, replay, and resume now point users toward agentflight handoff, while the handoff artifact itself tells users to share the generated local packet.

• AgentFlight now describes itself as a local-first review layer for coding agent sessions across package metadata, README, and product docs.

• Long suggested proof commands stay compact in high-density review surfaces while preserving the full suggested action where useful.

• Local AgentLoopKit evidence paths are filtered from AgentFlight changed-file review surfaces:

  • .agentloop/state.json
  • .agentloop/reports/**
  • .agentloop/handoffs/**
  • .agentloop/runs/**

• AgentLoopKit task contracts, policies, harness files, and gates remain visible for review.

• .projscan-memory/** remains suggestion-only guidance, not a built-in ignored path.

Fixed

• Reduced Review Intelligence and report noise caused by generated local workflow evidence.
• Split complex verification/profile and ProjScan-hint logic into smaller helpers without changing command behavior.
• Added regression coverage for PR-comment draft failure excerpts so stderr-preferred excerpts stay aligned with report/replay behavior.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• AgentLoopKit verification passed.
• ProjScan doctor passed with score 100/A.
• ProjScan preflight/review returned a documented manual release-signoff caution for scale risk only, with no concrete blockers.

AgentFlight v0.5.1 - 2026-06-17

Focused dogfood patch for v0.5.0 review artifacts.

Fixed

• Terminal agentflight verify now uses the same stderr-preferred failure excerpt as report/replay artifacts.
• Raw stdout/stderr evidence remains preserved.
• Reduced noisy AgentLoopKit Tooling diagnostics in Markdown reports.

Changed

• Trimmed long suggested proof commands in status, report, replay, and resume surfaces.
• Added shared output display helpers for compact proof commands and concise tool-report lines.
• Kept .projscan-memory/** as suggestion-only guidance, not a built-in ignored path.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• AgentLoopKit verification passed.
• ProjScan doctor passed with score 100/A.
• ProjScan preflight/review returned a documented manual release-signoff caution for scale risk only, with no concrete blockers.

[0.5.0] - 2026-06-16

Inline verification evidence and replay accessibility.

Added

• agentflight verify captures a short output excerpt (stderr preferred, otherwise stdout) at run time and stores it on the verification run.
• The replay shows the excerpt inline for failed runs, so a reviewer can see what broke without opening the evidence file. Passing runs keep the excerpt tucked inside the evidence details to stay calm.
• The Markdown proof report includes the failure excerpt as a fenced block for failed runs.

Changed

• Replay readout band now sticks to the top so risk and readiness stay visible while scrolling long records.
• Raised contrast on muted replay text and risk colors to meet WCAG AA for small text.

Added (replay)

• Print stylesheet for clean PDF handoffs and incident reconstruction: expands evidence, avoids mid-record page breaks, and preserves risk and readiness color.

Verification

• npm run verify passed.
• npm run format:check passed.

[0.4.2] - 2026-06-16

Replay UI redesign and refreshed product demos.

Changed

• Redesigned the HTML replay artifact as a flight-record evidence ledger: a verdict-led masthead where review readiness is the lead signal, an instrument-style readout band in place of the metric-card grid, a log-spine timeline whose event nodes encode type by shape and fill, and a verification ledger with explicit PASS/FAIL stamps.
• The replay now stays calm by default and applies semantic color only when risk is elevated or proof is failing, and never relies on color alone.
• Updated ProjScan to 4.5.0 and AgentLoopKit to 0.35.2.

Added

• Reproducible demo-asset pipeline: npm run demo:assets regenerates the replay screenshot and scroll GIF (Playwright) and the terminal workflow GIF (VHS) from scripts/demo/ and docs/marketing/.
• Refreshed README with the new terminal workflow GIF and an animated replay walkthrough.

Verification

• npm run verify passed.
• npm run format:check passed.

[0.4.1] - 2026-06-15

Review Intelligence trust patch focused on v0.4.0 dogfood findings.

Fixed

• Detect incomplete verification attempts when a verification_started event has no later completed result.
• Incomplete verification now appears as a blocking Review Intelligence proof gap and prevents Ready for review.
• Report output no longer shows legacy verification-gap text that conflicts with Review Intelligence proof gaps.
• .agentflight/config.json is classified as AgentFlight project config instead of generic unknown.

Changed

• agentflight verify emits a minimal heartbeat while long-running verification commands are still active.
• .projscan-memory/memory.json now triggers an informational suggestion to add .projscan-memory/** to changedFileFilters.ignore without hardcoding that path as a built-in ignore.

Documentation

• Documented incomplete verification handling, heartbeat output, and optional ProjScan memory filters.
• Updated v0.4.0 dogfood findings with v0.4.1 patch candidate status.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• AgentLoopKit verification passed.
• ProjScan doctor passed with health 100/100.
• ProjScan preflight returned proceed with no caution.
• Local packed-package smoke test passed.

[0.4.0] - 2026-06-14

Added

• Added deterministic Review Intelligence for coding agent sessions.
• Added review focus ranking to highlight the files developers should inspect first.
• Added proof gap detection for missing or failed verification evidence.
• Added clearer readiness states and next-best-action guidance.
• Added config-driven generated/internal changed-file filters via changedFileFilters.ignore.
• Added documentation for generated/internal file filters.
• Added completion audit for v0.4.0.

Changed

• status, report, replay, resume, and snapshot now include review intelligence where appropriate.
• Updated replay UI screenshot to show review focus, proof gaps, readiness, and timeline.
• Hardened malformed changedFileFilters.ignore handling.
• Kept .agentflight/config.json visible while runtime .agentflight/ artifacts remain filtered.

Deferred

• PR comments.
• JSON/CI integration.
• ProjScan-enriched ranking.
• Heartbeat/progress output for long verification.
• Interrupted verification cleanup.
• Tool availability messaging alignment.
• Cloud/login/billing/Pro/Team/GitHub App.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• AgentLoopKit verification passed.
• ProjScan doctor passed with health 100/100.
• ProjScan preflight returned a documented scale/complexity caution that was manually reviewed and accepted.

[0.3.3] - 2026-06-14

Patch candidate focused on dogfood findings from published agentflight@latest v0.3.2.

Changed

• Improved HTML replay readability with a calmer developer-review layout, clearer timeline rows, cleaner file grouping, and denser verification evidence rows.
• Moved replay stdout/stderr evidence paths into collapsed native details so proof stays available without dominating the page.

Fixed

• Excluded AgentFlight runtime artifacts from changed-file analysis by default:
  • .agentflight/sessions/**
  • .agentflight/reports/**
  • .agentflight/current/**
  • .agentflight/evidence/**
• Kept .agentflight/config.json visible in changed-file analysis because it is user-controlled project configuration, not runtime evidence.

Documentation

• Added PRODUCT.md with the confirmed product UI direction: precise, calm, trustworthy.
• Added docs/development/v0.3.2-dogfood-findings.md with dogfood results from AgentFlight, ProjScan, and fifa-predictor.
• Updated the development log with dogfood, UI polish, and patch verification evidence.

[0.3.2] - 2026-06-13

Changed

• Added the official AgentFlight logo to the README.
• Added the Baseframe Labs AgentFlight website link near the top of the README.
• Added an animated CLI workflow preview before the 60-second workflow.
• Added a "Watch The Flow" section explaining start -> verify -> snapshot -> status -> report/replay/resume.
• Updated package homepage to the AgentFlight website page.
• Expanded packaged assets so npm README logo and animation references resolve.

Documentation

• Updated launch notes with the AgentFlight logo reference.
• Updated development log with branding and README animation verification.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• ProjScan preflight passed.
• AgentLoopKit verification passed.

[0.3.1] - 2026-06-13

Changed

• Reworked the README around a clearer 60-second AgentFlight workflow.
• Added sample CLI output for status, report, replay, and resume.
• Added a replay timeline screenshot to make the v0.3.0 experience easier to understand.
• Added a basic example session walkthrough and v0.3.0 launch-note drafts.
• Improved npm keywords and package metadata.
• Narrowed packaged files to include useful README-linked docs/assets without shipping marketing drafts.

Documentation

• Added launch notes draft for v0.3.0 public/demo positioning.
• Updated development log with post-v0.3.0 polish verification.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• ProjScan preflight passed.
• AgentLoopKit verification passed.

[0.3.0] - 2026-06-13

AgentFlight now records session events and snapshots so reports and replays show how a coding session evolved.

Added

• Added agentflight snapshot --note "..." to record a local checkpoint for the active session.
• Added session-level events with backward compatibility for older sessions.
• Added event recording for session start, verification attempts, snapshots, report generation, replay generation, resume prompt generation, and doctor runs.
• Added timeline sections to Markdown reports and HTML replays.
• Added latest snapshot context to status and resume prompts.
• Added changed-file groups to replay output.
• Added tests for session events, snapshot creation, missing active sessions, timeline rendering, older session compatibility, and verification events.

Changed

• Replays now feel more like flight recorder timelines instead of final-state summaries.
• Reports now include a concise event timeline before changed files and proof evidence.
• Resume prompts now include latest snapshot notes and current verification state.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed for agentflight@0.3.0.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• ProjScan preflight passed with health 100/100.
• AgentLoopKit verification passed.

[0.2.1] - 2026-06-13

Patch release candidate focused on friction found while dogfooding the v0.2.0 core workflow.

Changed

• agentflight verify now prints the stdout and stderr evidence paths immediately after each recorded run.
• Failed verification output and downstream next actions now include the exact command to rerun.
• agentflight report, agentflight replay, and agentflight resume now avoid stale "generate a report" next actions once proof is ready.
• HTML replays now include a compact summary strip for risk, changed files, proof counts, and review readiness.
• AgentLoopKit workflow files under .agentloop/ are treated as low-risk dogfooding artifacts instead of unknown code changes.

Fixed

• ProjScan and AgentLoopKit adapters now prefer repo-local node_modules/.bin binaries before PATH-global commands, preventing stale global versions from appearing in reports.
• Tool adapter version output is normalized when CLIs include decorated version text.

Verification

• Targeted v0.2.1 regression tests passed.
• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed for agentflight@0.2.1.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• ProjScan preflight passed with health 100/100.
• AgentLoopKit verification passed.

[0.2.0] - 2026-06-13

AgentFlight now captures real verification evidence and uses it across status, report, replay, and resume.

Added

• Added agentflight verify to run verification commands and capture local evidence.
• Added verification run persistence in session records, including command, timestamps, duration, exit code, status, stdout path, and stderr path.
• Added .agentflight/evidence/ for local stdout/stderr artifacts.
• Added evidence-aware status, report, replay, and resume outputs.
• Added tests for verification success, failure, persistence, evidence-aware outputs, and v0.1 session compatibility.

Changed

• agentflight status now reports changed areas, proof gaps, review readiness, and a next action based on captured evidence.
• agentflight report now includes verification evidence and honest review recommendations.
• agentflight replay now renders verification cards in the local HTML artifact.
• agentflight resume now includes verification gaps, the exact next command, and stronger continuation guardrails.

Fixed

• agentflight doctor now checks write permission for .agentflight/, not just path existence.
• agentflight --version now reports the package version instead of the stale 0.1.0 value.

Verification

• npm run verify passed.
• npm run format:check passed.
• npm pack --dry-run passed.
• npm audit --audit-level=moderate found 0 vulnerabilities.
• ProjScan preflight passed.
• AgentLoopKit verification passed.

[0.1.1] - 2026-06-13

Fixed

• Fixed npm .bin symlink invocation so npx agentflight --help runs the CLI instead of exiting silently.

Added

• Added CI verification on pushes and pull requests.
• Added tag-based npm publishing through GitHub Actions Trusted Publishing.
• Added release process documentation.

[0.1.0] - 2026-06-13

Added

• First AgentFlight MVP CLI.
• Added init, start, status, report, replay, resume, and doctor commands.
• Added local .agentflight/ config, session, report, replay, and resume prompt artifacts.
• Added ProjScan and AgentLoopKit adapters with graceful fallbacks.
• Added TypeScript, Vitest, ESLint, Prettier, and npm package setup.
• Added README, architecture docs, dogfooding docs, verification docs, roadmap, and monetisation notes.