Changelog

[1.7.0] - 2026-08-14

Added

Global skills — ck install --global, ck update --global, ck check --global, and the new ck uninstall --global — install the squad pipeline (10 skills) and spec family (spec, spec-component) once to ~/.claude/skills/, shared across every project instead of duplicated per-repo via ck install. Each skill is fully self-contained (source command content is inlined at generation time — no ~/.contextkit/ mirror to manage), and the two skills with hard local-.contextkit/-dependencies (spec's existence guard, spec-component's template file) are adapted for the global variant so they work in any repo with zero local setup. ck update --global regenerates/overwrites in place. Every generated file carries a contextkit-managed/contextkit-version marker in its frontmatter, so ck uninstall --global can never remove a same-named skill you wrote yourself, and ck check --global can report version drift without needing separate state. Project-level .claude/skills/ (if present) still takes precedence over the global copies, so existing per-project installs are unaffected.
ck uninstall — new top-level command (currently --global only; per-project uninstall of standards/CLAUDE.md/hooks is a separate future effort). Safe by default: a bare ck uninstall --global lists what's installed instead of removing anything — you need an explicit skill/family name or --all.

Removed

review, refactor, test, and doc skills — dropped from all three skill-generating platforms (Claude Code, Codex, Cursor). review/test were largely superseded by the squad pipeline's squad-review/squad-test phases for anyone running squad end-to-end; doc overlapped with the more precise doc-arch/doc-feature/doc-component (3-level docs) and squad-doc. analyze is unaffected and still generates .contextkit/standards/. Existing installs get these cleaned up automatically on the next ck update (added to the legacy-file cleanup list) — the removed command files remain in git history if ever needed.

[1.6.1] - 2026-08-11

Fixed

ck update now syncs status.json's version field, not just config.yml's — status.json was only ever written once at ck install time; StatusManager.updateVersion() existed but was never called from the update flow, so status.json's version (and ck status's own independent update-check, which reads it) drifted permanently stale after every subsequent ck update, while config.yml's version stayed correct.
Pre-push quality gate failures now actually show the failure banner — hooks/pre-push set trap on_gate_failure ERR but was missing set -o errtrace, so the trap never fired for a failure inside a function (which every real gate runs through: run_gate, run_node_gates_core, etc.). A failing gate blocked the push correctly but silently — no "❌ Quality Gates FAILED" banner, no indication of what happened or where. Also fixed the summary box's Gates:/Total: lines, which used hardcoded padding and could render with a missing or misaligned border; they're now printed as plain lines below the box instead.
ck update/ck status no longer leak literal quote characters from single-quoted config.yml values — their hand-rolled YAML parsers only stripped double quotes (.replace(/"/g, '')), so a version: '1.0.0'-style (single-quoted) config produced messages like Updating from '1.0.0' to 1.6.0. Beyond the display glitch, the mangled string fed into isNewerVersion()'s numeric comparison, where corrupted segments parse to NaN — a real (if edge-case) correctness bug, not just cosmetic. Also fixed status.js's version: match missing the top-level-only guard update.js already had, which could pick up the indented _source.version depending on field order.

[1.6.0] - 2026-08-11

Changed

/squad-spec and /squad-auto commit per story/task, not just at the end of a run — right after a story's (or task's) Doc phase passes, the orchestrator stages exactly the files named in that unit's Changes Made/Tests Written/Doc Notes (not a blanket git add -A) and commits in Conventional Commits format, before moving on. Skips cleanly if there's nothing to commit or the project isn't a git repo; if anything unrelated shows up staged, or the commit itself fails (e.g. a pre-commit hook), it stops and surfaces that instead of continuing with uncommitted work. Keeps history bisectable and revertable per story instead of one scope-sized commit at the end.
/squad-spec Architect now resolves blocking open questions itself — a new ### Open Question Resolutions section in the handoff template requires Architect to record the assumption made and its rationale for each open question a story's spec flags as blocking, instead of silently deciding and leaving it buried in prose. Scoped to /squad-spec specifically, since (unlike /squad//squad-auto, which can pause via arch-clarify) it runs unattended with no human to escalate to mid-run.

[1.5.0] - 2026-08-10

Changed

/squad-spec dispatches each phase as an isolated subagent — Architect, Dev, Test, Peer Review, Review, and Doc each now run in a fresh Task-tool context on Claude Code, reading only the handoff file rather than the accumulated reasoning of earlier phases. Prevents Review from anchoring on Dev's own account of its work. Codex falls back to the previous inline-in-context behavior — no equivalent subagent-spawn primitive is available to a /goal-driven skill yet.
/squad-spec Review and Peer Review default to claude-haiku-4-5-20251001 — both are verification passes against an already-written checklist (acceptance criteria, prior Issues Found), not design work, so they route to the cheaper model by default. Architect and Dev stay on the primary model.
/spec writes acceptance criteria in EARS format — SPEC.md's ## Stories section now has a dedicated ### Acceptance Criteria block per story (WHEN [event], the system SHALL [response] / IF [precondition], THEN the system SHALL [response]), instead of folding criteria into the one-line /squad command string. /squad-spec's handoff pre-fill now copies this block verbatim rather than re-deriving criteria, and Review checks each line individually against the diff. /spec --extend appends new stories' AC blocks alongside their table rows.
/squad-spec Step 5 full-suite gate runs in the foreground with low/serial parallelism — replaces the previous backgrounded, max-worker run. The background+parallelism combo was pinning the machine's CPU while barely saving any real time (it only overlapped the trivial summary-writing step, not actual work), and the gate already blocked until the suite finished either way. Advancing to the next scope still waits for a clean result.

Fixed

squad-spec skill was missing Task in allowed-tools — would have silently blocked the subagent dispatch above from working at all.

[1.4.0] - 2026-08-05

Changed

/squad-spec test scope — per-story Test phase now runs only tests touching that story's changed files; the full suite runs once per scope, at the new Step 5 regression gate, backgrounded so it doesn't block reporting.
/squad-spec review failures now self-repair — a needs-work verdict sends the story back to Dev automatically (up to 2 retries, tracked via a new attempts field) with an independent Peer Review pass on each retry, before finally stopping for a human.
/squad-spec Step 3 resume logic — explicit precedence for needs-work stories (blocks only their dependents, not the whole run) and a defined recovery path for stories interrupted mid-phase (resumes from that phase instead of undefined behavior).

[1.3.0] - 2026-07-24

Removed

Claude Code PostToolUse format+lint hook — ck install/ck update no longer writes a PostToolUse hook to .claude/settings.json. It ran the project's format+lint command after every single Edit/Write during a Claude Code session, which noticeably slowed sessions down on any file-heavy task. Quality is still enforced at push time via the existing pre-push git hook. Removed lib/utils/hook-detector.js and lib/utils/claude-settings.js along with their tests, since nothing else used them. Existing installs keep whatever hook entry they already have in .claude/settings.json (it's user-owned, gitignored state) — remove it by hand if you had one; ck update will not re-add it.

[1.2.3] - 2026-07-22

Added

lib/utils/standards-migrations.js — registry of content fixes to generic sections of already-installed standards/*.md files. ck update never re-downloads ai-guidelines.md/architecture.md/code-style.md (they're user-owned so /analyze customizations survive), which meant a wording/behaviour fix to shipped boilerplate never reached existing installs. update.js#applyStandardsMigrations now replaces each entry's exact from text with to when found verbatim; if the section was hand-edited (by the user or /analyze) and no longer matches, the entry is recorded in a new pending_standards_updates list in config.yml instead of being force-applied. See standards-migrations.md.
standards-aware skill reconciles pending standards updates — before doing agentic work, it now checks config.yml's pending_standards_updates list and, for each entry, proposes folding the upstream change into the project's customized content rather than leaving it stale indefinitely.
pending_standards_updates config.yml field — added via config-schema.js (existing installs) and install.js's fresh-install template.

Changed

ai-guidelines.md's "README.md Context Discovery" rule scoped down — now skips the check for quick lookups, one-line fixes, and single-file tasks with an already-clear purpose, keeping the full check only for substantial feature/component work. This is the first entry in the new standards-migrations.js registry above.

[1.2.2] - 2026-07-21

Changed

/spec and /squad-spec output moved to .contextkit/spec/ — previously written to a root-level spec/ folder that was always committed. Per-feature specs are ephemeral working state (deleted/archived once the stories are implemented via squad), so the folder now lives under .contextkit/ and is gitignored/archived the same way .contextkit/squad/ already is (.contextkit/spec-archived-[timestamp]/ on conflict). Updated commands/spec/spec.md, commands/squad/squad-spec.md, the spec skill's description text, README.md, and the public docs site's /docs/spec page. lib/commands/install.js now adds .contextkit/spec/ and .contextkit/spec-archived-*/ to generated .gitignore entries.

Fixed

ck update no longer risks writing the project version into the wrong config.yml field — parseConfig and updateConfigVersion matched any version: key, including the indented _source.version field (which tracks the CK tool version, not the project's). Both are now anchored to the top-level, unindented version: line only.

[1.2.1] - 2026-07-14

Fixed

ck update no longer drops config.yml blocks it doesn't recognize — restoreUserConfig used to fully regenerate config.yml from a narrow hardcoded template, silently discarding any block absent from that template (response_style, required/optional/conditionals, analysis_scope/analyzed_packages, features.squad_ci_workflow, profile, metadata, etc.). Replaced with appendMissingSettings, which diffs the raw file against a new config-schema.js registry and only appends whatever's missing — with an explanatory comment — leaving everything else untouched. This also restores check.js's required/optional-file gate and analyze.js's monorepo scope memory, both of which were being silently disabled on any project that had run ck update even once.

Added

lib/utils/config-schema.js — single registry of additive config.yml settings, consumed by both a future install.js template pass and update.js#appendMissingSettings. See lib/utils/config-schema.md.

Docs

migrations.md — clarified when to use format-version migrations vs. the new additive config-schema.js registry.

Tests

update unit tests — added tests 22–26 covering appending response_style, required/optional/conditionals/analysis_scope, inserting features.squad_ci_workflow into the existing features: block, not duplicating settings that already exist, and not reordering/removing any pre-existing line.

[1.2.0] - 2026-07-13

Added

Configurable response style — new response_style block in config.yml with two independent per-project flags: chat_minimal_words (terse, plain-text chat explanations) and diagrams_in_docs (allow Mermaid in generated .contextkit/standards/*.md). Both default to the existing compact behavior (true/false). Checked by ai-guidelines.md, /analyze's Output Density rules, and the compress skill.
Generated CLAUDE.md now points to the response_style config block so the toggle is discoverable every session, not just when ai-guidelines.md loads via the standards-aware skill.

Tests

install unit tests — added tests 64–65 verifying config.yml ships the response_style defaults and ai-guidelines.md documents both flags.
claude-integration unit tests — added test 19 verifying generated CLAUDE.md references response_style / config.yml.
download-manifest test — updated test 26 for the reworded (now config-aware) Output Density bullet in analyze.md.

[1.1.7] - 2026-07-09

Changed

Always-loaded standards footprint cut ~6x — Claude Code and Codex now always-load only code-style.md and architecture.md via CLAUDE.md/AGENTS.md @imports. testing.md, workflows.md, ai-guidelines.md, glossary.md, decisions.md, roadmap.md, and corrections.md now load on demand via the test, agent-push-checklist, standards-aware, and new product-context skills instead. See docs/token-efficiency.md for the rationale.
analyze generates compact output by default, and the doc-* skills gained a Length Check split-proposal step.

Added

compress skill — retrofits oversized standards/docs files for information density; marker-safe on bridge files and backs up hand-edited skill files first.
product-context skill — loads architecture decisions and roadmap context on demand before proposing changes.
Update nudges — a throttled pre-push hook check and a self-update offer surfaced via standards-aware, so existing installs that never check for updates still get pointed at ck update.

[1.1.6] - 2026-06-29

Added

/squad-spec auto mode — running /loop /clear /squad-spec without a scope slug now processes ALL completed spec scopes in sequence. When a scope finishes it writes the next scope slug to .contextkit/squad/next-scope.txt and lets the loop continue rather than stopping. Only halts when all scopes are exhausted, printing a full cross-scope summary. Previously required manually re-triggering the loop for each scope.

[1.1.5] - 2026-06-29

Fixed

ck update resilience (complete) — ALL file downloads (commands, hooks, standards, types, scripts) are now non-fatal. Any single file failure logs a ⚠ Skipped warning and the update continues. Previously only types and legacy scripts were graceful; a transient 400 on any command file still aborted the entire update and rolled back the backup.
ck update skipped-file reporting — spinner now shows Files updated (N skipped — retry if issues persist) when downloads were skipped, making partial updates visible.

Tests

update unit tests — added tests 20–21 verifying that a 400 on a command file download is non-fatal and logs a warning instead of aborting.
CLI integration tests — expanded cli.test.js from 9 to 18 numbered tests; now covers ck update, ck check, ck analyze, ck note, ck run at the CLI level, and verifies ck update exits cleanly when ContextKit is not installed.
download-manifest test — updated URL regex to match refactored ${base} variable alongside original ${this.repoUrl}.

[1.1.4] - 2026-06-29

Fixed

ck update resilience — types files (strict.tsconfig.json, global.d.ts, type-check.sh, typescript-strict.json) and scripts/update.sh are now downloaded with graceful fallback. A transient network error on any of these non-critical files previously aborted the entire update and rolled back the backup. They now log a warning and continue.
squad-config integration test — fixed test paths that expected squad commands at .contextkit/commands/squad.md (flat) instead of the correct .contextkit/commands/squad/squad.md (subdirectory).

[1.1.3] - 2026-06-29

Added

/spec OVERVIEW.md — pass an overview file directly to start a new spec session. If an existing spec is detected from a different source file, prompts before archiving (spec/ → spec-archived-[timestamp]/). Same source resumes normally with no prompt.
/spec --add [scope-slug] — append a new scope to an in-progress spec and run it immediately. No need to manually edit PROGRESS.md.
/spec --extend [scope-slug] — add new stories to an already-written scope. Appends to the Stories table and Squad Commands block; leaves data model, API contracts, and UX flows untouched unless strictly required.
/spec loop termination — Step 5 now emits a stop signal so /loop /clear /spec terminates cleanly when all scopes are done (overnight batch runs).
/squad conversation extraction — running /squad with no explicit task args (or with a directive phrase like "let's do it", "let's go", "go ahead") now scans the current conversation for agreed tasks, shows a confirmation list, and proceeds. Previously required explicit task descriptions.

Changed

/squad-go — simplified to "squad + auto-run": follows the full /squad workflow (including conversation extraction) then immediately runs the pipeline with no checkpoint pause. Previously duplicated squad logic; now delegates to squad.md with checkpoint: none.
/squad-go SKILL.md — removed context: fork. Currently inert (fork is not yet implemented in Claude Code) but future-proofs the skill: when fork lands, leaving it would break conversation access.

Improved

ai-guidelines.md + AGENTS.md — added "Working on This Package" rule: when editing commands or skills in this repo, all mirror locations must be updated together (.contextkit/commands/ + commands/, .claude/skills/ + .codex/skills/).
.claude/skills/spec/SKILL.md — updated argument hint to include new flags; replaced stale "multi-round domain experts" description with accurate single-pass CTO description; removed context: fork.
.codex/skills/spec/SKILL.md — updated description to match new flags and behaviour.

[1.1.2] - 2026-06-25

Removed

/squad-auto-parallel — removed. Parallel subagents have no write coordination; tasks in a real batch almost always touch shared files (e.g. lib/index.js, README.md), producing conflicts that cost more to resolve than the parallelism saves. Use /squad-auto for all batch runs.

[1.1.1] - 2026-06-24

Changed

Codex integration — full skills support — ck codex now creates .codex/skills/ with 24 skill files (same set as Claude Code): spec, squad, squad-auto, squad-spec, squad-go, all squad phase commands, doc family, dev commands, and agent commands. Each skill reads the corresponding .contextkit/commands/ file. Previously only created AGENTS.md with no skill support.

[1.1.0] - 2026-06-24

Changed

/spec — complete redesign to single inline CTO pass — replaced the multi-agent pipeline (4 parallel domain experts, CTO challenges, revision round, 8+ agent spawns per scope) with a single inline CTO pass. One SPEC.md per scope instead of 7 intermediate files. Output now includes a ### Squad Commands section with copy-paste /squad commands for every story. Significantly lower token cost and simpler output.
/spec output format — SPEC.md now contains: Summary, Personas, Data Model (full schema + ERD), API Contracts (all endpoints), UX Flows (key journeys + edge cases), Stories table with sizes and dependencies, Squad Commands, Open Questions, Out of Scope. The ASSUMPTION: inline marker flags inferred detail for CTO review.
Removed spec-ux.md, spec-data.md, spec-systems.md, spec-planner.md — domain agent sub-commands no longer needed.

Added

/squad-spec [scope] — new command that loads a completed spec scope and runs the full squad pipeline (architect → dev → test → review → doc) for every story automatically. PO phase is skipped — the spec already is the PO spec. Designed for use with /loop /clear /squad-spec [scope] to process one story per loop iteration with a fresh context each time, preventing context bloat across long runs.

[1.0.3] - 2026-06-24

Changed

/spec pipeline — collapse sequential agents into orchestrator — the four single-sequential CTO agents (Scoper, Briefer, Challenger, Author) are now inline steps run by the orchestrator directly. Only the two parallel domain expert rounds (Round 1 initial, Round 3 revision) still spawn sub-agents. Agent count per scope reduced from 11–12 to 8. The spec-init.md, spec-brief.md, spec-challenge.md, and spec-author.md files are removed — their logic lives in spec.md.

[1.0.2] - 2026-06-20

Changed

Cursor /spec command — now points to the project-level spec pipeline (commands/spec/spec.md) with a description of the full multi-scope pipeline. Previously pointed to commands/dev/spec.md (the component spec).
Cursor /spec-component command — new generated prompt for the component-level MD-first spec workflow (split from the old /spec). Added to generatedFiles, contextkit-components.mdc rule, and showUsage() output.

[1.0.1] - 2026-06-08

Added

README.md context discovery rule — standards/ai-guidelines.md template now includes a ## README.md Context Discovery section instructing AI agents to check for README.md at the feature/page and component levels before starting work.
Squad Architect step 5.7 — /squad-architect now discovers and reads context files at all relevant documentation levels before exploring the codebase, so plans are informed by existing docs.
Standard filename guidance — standards/architecture.md template now documents README.md as the standard filename for feature/page and component levels, and named files (e.g. docs/architecture.md) for the architecture level.

[1.0.0] - 2026-06-08

Removed

vibe-kit / vk CLI aliases — removed from package.json#bin. Use contextkit or ck instead. The deprecation warning has been in place since 0.16.x; 1.0.0 is the hard cutover.

Fixed

Roadmap template placeholders — the roadmap.md generated by ck install no longer contains confusing raw brackets ([PHASE_NAME], [FEATURE], [ITEM]). Replaced with instructional HTML comments that guide users to fill in their own content.

[0.21.0] - 2026-06-08

Security

Next.js / React CVE fix — updated contextkit-docs dependencies (next, react-server-dom-webpack, react-server-dom-parcel, react-server-dom-turbopack) to patched versions resolving React Server Components vulnerabilities. Main package unaffected.

[0.20.0] - 2026-05-26

Removed

/fix command — removed across all platforms (Claude Code, Cursor). Use /squad for bug fix workflows instead.

Changed

Tagline updated to "Context Engineering + Agentic AI Pipelines" — README hero, website (hero, SEO metadata, footer, docs intro), and package.json now reflect both pillars equally.
package.json keywords — added agentic-pipeline, agentic-ai, squad-workflow, ai-agents, autonomous-agents, multi-agent, claude-code.

[0.19.0] - 2026-05-17

Added

/spec project specification pipeline — multi-round, multi-persona pipeline that turns a product overview into an implementation-ready spec before any code is written. Runs CTO Scoper → CTO Brief → parallel domain experts (UX, Data, Systems, Planner) → CTO Challenger → parallel revisions → CTO Author with final SPEC.md synthesis. Scope-by-scope with spec/PROGRESS.md checklist.
9 new spec sub-commands — spec-init, spec-brief, spec-ux, spec-data, spec-systems, spec-planner, spec-challenge, spec-author orchestrated by the main spec command.
/spec-component — renamed from the previous /spec command; writes a component-level MD-first spec colocated with source.

Changed

/spec skill — now points to the new project spec pipeline (spec/spec.md); component spec moved to /spec-component.

[0.18.0] - 2026-05-17

Added

/squad-go express command — single-invocation alternative to /squad + /squad-auto. Reads tasks from the current conversation, presents a confirmation list, writes PO specs, and immediately runs the full pipeline with no checkpoint pause.

[0.17.0] - 2026-05-17

Added

PostToolUse hook — ck install claude now writes a PostToolUse hook to .claude/settings.json that runs format+lint after every file edit in a Claude Code session. Auto-detected for Node.js (npm/pnpm/yarn/bun), Go, and Python. Gracefully skipped when no tooling is found.

[0.16.4] - 2026-04-10

Fixed

CI: npm auth — restore registry-url and NODE_AUTH_TOKEN via NPM_TOKEN secret; --provenance handles attestation only, not authentication

[0.16.3] - 2026-04-10

Fixed

CI: remove registry-url from setup-node — registry-url writes an .npmrc expecting NODE_AUTH_TOKEN which conflicts with OIDC Trusted Publishing; removing it lets --provenance handle auth correctly

[0.16.2] - 2026-04-10

Fixed

CI: npm auth — add NODE_AUTH_TOKEN to publish step so the registry accepts the NPM_TOKEN secret

[0.16.1] - 2026-04-09

Fixed

CI: npm self-update removed — npm install -g npm@latest was corrupting the npm installation in the GitHub Actions hostedtoolcache; Node 22 ships with npm 10+ which already supports --provenance

[0.16.0] - 2026-04-09

Added

Smart git hooks detection — ck install detects existing hooks managers (Husky, Lefthook, simple-git-hooks, existing core.hooksPath, scripts in .git/hooks/) and suggests integration instead of overriding
ck install --force — regenerate all files including user-customized standards
ck update --force — also regenerate user-owned files (standards, glossary)
/squad standalone mode — if .contextkit/ isn't set up, /squad offers to create just .contextkit/squad/ so the pipeline works without a full ck install; never creates .squad/ in the project root

Improved

ck install smart reinstall — skips user-owned standards files on reinstall; prints skip message with --force hint
ck update non-destructive — never overwrites user-owned files (glossary.md, standards); preserved files logged clearly
ck claude / ck cursor version-aware — skip regeneration when integration is already up to date; regenerate when version mismatches or files are missing
ck analyze tech stack fix — no longer generates JS/TS standards for Go, Python, Rust, Java, and other non-JS projects
Gitignore taxonomy — addContextKitGitignoreEntries writes 6 targeted runtime-state entries under a descriptive header; detects and handles existing ignore-all .contextkit/* pattern gracefully; upgrades legacy # ContextKit header
/doc-arch dynamic output — output filename derived from topic argument, PR title, or branch name (docs/<topic>.md); confirms filename when inferred, falls back to docs/architecture.md

[0.15.1] - 2026-04-07

Improved

Squad pipeline: sequential task execution enforced — /squad-auto writes each file to disk before moving to the next task
Testing standards: Testing Trophy levels — testing.md updated with unit/integration/e2e decision table and change-driven test selection rules

[0.15.0] - 2026-04-07

Added

Skills format — Claude commands migrated to skills format with rich frontmatter; Cursor rules updated to match
ck install gitignore entries — adds .contextkit/squad/ and .contextkit/squad-done-*/ to .gitignore on install

Fixed

ck update cursor paths — corrected path resolution for Cursor rules on update

[0.14.1] - 2026-03-24

Fixed

.claude/commands/ path bug — all generated slash command wrappers now reference correct .contextkit/commands/ subdir paths (dev/, squad/, docs/, agents/)

Added

New slash command wrappers: /agent-push-checklist, /context-budget, /standards-aware
ck install now appends .claude/settings.local.json to .gitignore automatically (no-op if file absent or entry already present)

[0.14.0] - 2026-03-23

Removed

ck gates command — gate configuration is managed directly via .contextkit/quality-gates.yml; editing the file is simpler and more transparent than a CLI wrapper

[0.13.6] - 2026-03-17

Added

Quality Gates: format script gate — if package.json has a format script, runs it in write mode then blocks the push if any tracked files changed, with a remediation message (git add -A && git commit -m 'chore: format')
Quality Gates: lint script gate — if package.json has a lint script, runs it and blocks on non-zero exit
Both gates respect quality-gates.yml (disable: [format] / disable: [lint])
ck install: quality tooling scaffold — when the pre-push hook is enabled on a Node.js project with no format/lint scripts, prompts to scaffold a minimal prettier + eslint setup (adds scripts, .prettierrc, .prettierignore, eslint.config.js, and installs devDependencies using the detected package manager); default No; skipped in CI/non-interactive mode; respects existing configs

Improved

ContextKit own tooling — added prettier + eslint (flat config) as devDependencies; npm run format and npm run lint scripts now active so the gates fire on ContextKit development too; initial format pass applied to all source files

Fixed

analyze.js — const hasBackend reassigned in loop; changed to let (surfaced by eslint)
install.js — two useless escape sequences in shell template literal (\$@ → $@, \$( → $()

[0.13.5] - 2026-03-16

Added

Quality Gates: 3 new stacks — Kotlin (ktlint + Gradle test), Swift (SwiftLint + swift test), .NET/C# (dotnet build + dotnet test)
Quality Gates: config file — .contextkit/quality-gates.yml with disable: list lets you opt out of specific gates without editing hook code (e.g. disable: [eslint, prettier])
Quality Gates: per-gate timing — each gate shows elapsed time (✓ 2s) after completion
Quality Gates: DRY_RUN=1 mode — set DRY_RUN=1 git push to preview which gates would run without executing them; exits 0
Quality Gates: monorepo support — npm/pnpm workspace projects scope gates to affected packages when all push changes are inside workspace package directories

Improved

Quality Gates: docs — failure banner callout with real output example; commit-msg section adds auto-skip note, 10-char minimum, improve example; Node.js table row adds e2e + package.json note; workflow block replaced with numbered list; "skipped silently" used consistently; Troubleshooting section added; hooks-copied-at-install clarified; Kotlin/Swift/.NET rows added to framework table

[0.13.4] - 2026-03-16

Added

/doc-arch command — generates docs/architecture.md (Level 1: system boundaries, key flows, Mermaid diagrams). Stack-aware; can target current branch or a PR number.
/doc-feature command — generates docs/features/<name>.md (Level 2: feature scope, components, data flows, sequence diagrams). Accepts name, directory, or PR number.
/doc-component command — generates a colocated <name>.md next to a file (Level 3: props/API, usage example, edge cases). Accepts file path or directory.
Docs: md-first page — commands section with 3-level hierarchy table, per-command subsections, stack-awareness note, when-to-use-which table, squad integration note
Docs: slash-commands page — doc-arch, doc-feature, doc-component entries added

[0.13.3] - 2026-03-16

Added

CI Squad — label any GitHub issue squad-ready to trigger a full squad pipeline in GitHub Actions (PO → Architect → Dev → Test → Review → Doc) and open a draft PR automatically. No local setup required.
ck install CI Squad prompt — opt-in prompt during install downloads squad-issue.yml to .github/workflows/ and records squad_ci_workflow: true in config
ck update CI Squad sync — re-downloads squad-issue.yml on ck update when squad_ci_workflow flag is set
squad-ci.md command — new command distributed via install/update; instructs Claude Code CLI to run the full pipeline non-interactively in CI mode. On clarify: posts a GitHub comment and exits 0. On completion: writes ci-result.md for the PR body.
templates/github-actions/squad-issue.yml — new GitHub Actions workflow template with branch creation, clarify comment flow, concurrency guard, and draft PR creation
Docs: CI Squad page — new website page covering setup, how it works, writing good issues, clarify flow, and behaviour table

Fixed

Website: commit types — quality-gates page was showing stale style type; updated to improve (matches CLI and standards)

[0.13.2] - 2026-03-13

Fixed

commit-msg hook — length check now tests the subject line only (head -n1), not the full message including body. A short subject with a long body previously passed incorrectly.
commit-msg hook — length check now runs before the format check (was unreachable dead code — any message passing the format regex was already >10 chars).
pre-push hook — fixed variable-width content inside box borders; Project type and success summary lines moved outside the box to prevent broken border alignment.

Added

pre-push hook — ERR trap prints a clear ❌ Quality Gates FAILED — push blocked. banner when any gate fails, replacing silent exit.

Changed

commit-msg hook — removed style from allowed commit types (not used in this project; was inconsistent with documented types in workflows.md and ai-guidelines.md).
README — commit types list updated to remove style, matching the hook and standards files.

[0.13.1] - 2026-03-07

Fixed

ck ai stale reference — removed leftover ck ai <cmd> line from post-install Quick Reference output (command was removed in 0.13.0)
Install "In CLI" example — replaced ck ai "..." example with generic slash command guidance

Added

OpenCode auto-detection — ck install now detects the opencode binary in PATH and suggests the OpenCode integration automatically, consistent with how claude, codex, and gemini CLI tools are detected

Changed

/squad vague task handling — PO now pushes back with up to 5 clarifying questions when a task description is too ambiguous to spec. Sets status: po-clarify, writes questions to the handoff, and pauses the pipeline. Running /squad again (no args) with answers resumes and writes the full spec.
/squad Clarification Mode — split into two paths: kickoff clarification (questions in PO Spec block → write spec, advance to architect) and downstream clarification (questions from Architect/Reviewer → existing behavior)

[0.13.0] - 2026-03-07

Removed

ck ai — removed AI chat fallback command; users interact via Cursor, Claude Code, OpenCode, etc.
ck dashboard — removed observability dashboard command
ck pull / ck publish — removed local registry commands (no shared registry was implemented; feature deferred)
Unknown commands no longer silently forward to AI; now print an error and exit with code 1

Added

format_version field in .contextkit/config.yml — written as format_version: 1 on ck install
Migration runner (lib/utils/migrations.js) — ck update detects outdated format versions and migrates automatically; v0→v1 migration bootstraps the field on existing pre-1.0.0 installs

Changed

Node.js engine requirement bumped to >=18.0.0 (Node 14/16 are EOL)
improve added as a valid commit type in README (was already in workflows.md and the commit-msg hook)
Docs site — removed observability/dashboard sections from Commands and Enterprise pages; quick-start and docs README now require Node 18.x

[0.12.22] - 2026-03-06

Fixed

commit-msg hook — merge commits (Merge branch ...), revert commits, fixup, and squash commits are now skipped automatically instead of being rejected with a conventional format error.
commit-msg hook — added improve as a valid commit type (was already listed in ContextKit's own standards but missing from the hook regex).

[0.12.21] - 2026-03-06

Added

Architect split signal — the Architect now evaluates task complexity before writing the plan. If a task is too large (>~7 files, multiple independent concerns, or significant wasted effort risk), it writes a ### Recommended Split with proposed sub-tasks and sets status: po-clarify. The PO then approves the split (running /squad as a batch) or dismisses it and proceeds as-is — no code is written until the PO decides.

Changed

/squad Clarification Mode — detects ### Recommended Split in the Architect Plan block and guides the PO through the two response options (approve split or proceed as-is), separate from the standard spec Q&A flow.
/squad-auto po-clarify message — updated to mention the split recommendation scenario so users understand what to expect when running /squad.

[0.12.20] - 2026-03-06

Added

/squad-doc phase — new final pipeline step that runs after review passes. Creates companion .md files for every new code file and updates existing ones for significant modifications. Enforces MD-first as a quality gate rather than an afterthought.
commands/squad-doc.md — new command file distributed via ck install and ck update.
/squad-doc slash command — available in Claude Code after ck claude.

Changed

Squad pipeline flow — now architect → dev → test → review → doc → done. Review PASS routes to doc status instead of directly to done.
squad-auto — runs the Doc phase inline after each task's review passes.
squad-auto-parallel — runs a sequential Doc pass after all parallel agents complete.
Handoff template — includes new ## 7. Doc section.

[0.12.19] - 2026-03-06

Added

Passive update notifications — ck now checks npm for newer versions of @nolrm/contextkit once per 24 hours and prints a non-blocking one-liner after any command if an update is available. Suppressed automatically in CI environments.
ContextKit version stamp in CLAUDE.md — the generated CLAUDE.md now includes the installed version (Version: X.Y.Z) so AI tools can report it and check for updates when running /ck.
/ck update check step — the health-check command now runs npm view @nolrm/contextkit version, compares to the installed version in status.json, and adds an update row to the status table.

Changed

health-check.md now downloaded from GitHub (like all other command files) instead of being written inline at install time. Existing users get the updated file via ck update.

[0.12.17] - 2026-03-04

Changed

Banner art — replaced "CK" initials ASCII art with full "ContextKit" slant-style logo on ck install and ck update.

[0.12.15] - 2026-03-04

Added

/squad-reset — new slash command to clear .contextkit/squad/ and start fresh. Reports what was removed and flags any in-progress tasks before deleting.

Changed

/squad mixed-state handling — instead of hard-stopping when both handoff.md and manifest.md exist, /squad now offers to reset inline and continue with the provided task. Falls back to instructing /squad-reset if no task was given or the user declines.

Removed

/squad-peer-review removed from the pipeline. Existing installs have the file deleted automatically on ck update.

Fixed

ck update legacy cleanup — removes .contextkit/commands/squad-peer-review.md on update.

[0.12.14] - 2026-03-04

Changed

/squad-run renamed to /squad-auto, /squad-run-agents renamed to /squad-auto-parallel — "auto" unambiguously signals hands-free execution; "run" was ambiguous (run what? run a step?). This is a breaking rename for muscle memory — existing users should note the new names.
Squad command menu labels — added contextual suffixes so the slash-command dropdown is self-explanatory without guessing: (start here), (recommended), (manual step 1/4) through (manual step 4/4).
Post-kickoff message — /squad now recommends /squad-auto as the primary next step (hands-free) and lists manual steps as a secondary option. Previously it only mentioned /squad-architect.

Breaking

/squad-run no longer exists. Use /squad-auto instead.
/squad-run-agents no longer exists. Use /squad-auto-parallel instead.

[0.12.13] - 2026-03-04

Added

.contextkit/README.md — generated on ck install. Explains what the directory is, includes the install command (npm install -g @nolrm/contextkit) and npm link so any developer who encounters the folder knows what manages it.
_source block in .contextkit/config.yml — machine-readable attribution written at install time with tool, version, and npm fields.
Updated attribution marker — the  comment in all bridge files now includes the npm package name and link: . Backward-compatible: existing installs with the old short marker are detected and upgraded on next write.

[0.12.12] - 2026-03-03

Changed

/squad-batch removed — merged into /squad. Pass one task for single-task mode, two or more for batch mode. /squad auto-detects and routes accordingly.
commands/squad.md, .contextkit/commands/squad.md — full rewrite combining single-task, batch, append, and clarification flows into one command. Adds mixed-state detection, po in-progress warning before overwrite, done archiving for both single and batch, and standardised handoff template with Visual Assets and User Clarifications sections.
lib/integrations/claude-integration.js — removed .claude/commands/squad-batch.md from generated files; added to legacy cleanup list so existing installs remove the stale file automatically.
lib/integrations/cursor-integration.js — same as above for .cursor/prompts/squad-batch.md.
lib/commands/install.js, lib/commands/update.js — removed squad-batch.md download.
README.md, contextkit-docs — updated all batch flow examples to use /squad.

Breaking

/squad-batch no longer exists as a command. Use /squad "task 1" "task 2" ... instead.

[0.12.11] - 2026-03-03

Changed

commands/squad.md, .contextkit/commands/squad.md — PO agent now checks for screenshots or images the user attached to the /squad invocation. If present, each image is saved to .contextkit/squad/assets/ and listed with a one-line description under the new optional ### Visual Assets section in the handoff. If no images are provided the section is left empty and the pipeline continues unchanged.
commands/squad-architect.md, .contextkit/commands/squad-architect.md — Architect now reads any ### Visual Assets file paths from the PO Spec before writing the plan.
commands/squad-dev.md, .contextkit/commands/squad-dev.md — Dev now reads any ### Visual Assets file paths from the PO Spec before implementing.
README.md — Added Visual Assets section to the Squad Workflow docs and updated the PO role description.

[0.12.9] - 2026-03-03

Changed

lib/commands/install.js — When a user declines reinstall, the CLI now shows a hint: run ck update to get the latest command and squad files without overwriting customized standards.

[0.12.7] - 2026-03-01

Changed

commands/squad-batch.md — Batch kickoff now detects whether a manifest already exists. Fresh run creates a new manifest and handoff files as before. If a manifest exists, append mode activates: new tasks are numbered from where the existing batch left off, the manifest total: is updated, and PO specs are written for new tasks only — existing handoffs are never re-processed. Run /squad-run after to continue the full pipeline.

[0.12.6] - 2026-03-01

Changed

Standards template — standards/glossary.md is now a neutral project-agnostic template ([YOUR_DOMAIN] / [term] placeholders) instead of shipping with e-commerce/accounting examples that don't apply to most projects.
commands/analyze.md — Analysis now includes a Term & Glossary Detection step: AI scans the codebase for domain-specific vocabulary and appends discovered terms to glossary.md. Added Update Strategy rule: placeholder files are replaced, files with existing custom content are improved and augmented — never overwritten. Added ai-guidelines.md, workflows.md, and glossary.md to the list of files updated during analysis.
standards/README.md — Quick Reference section is now stack-neutral (removed React/TypeScript-specific "functional components" and "atomic design" advice). Added workflows.md to the Standards file list. Updated glossary description to "Project terminology and domain glossary".
Root standards/ reference files — Added clear banner to code-style.md, testing.md, architecture.md, ai-guidelines.md, and workflows.md clarifying they are reference examples (not auto-installed) showing what ck analyze can generate.
Docs & README — Updated glossary description from e-commerce-specific examples to generic domain terminology. Updated quick-start and platform-examples usage snippets to reflect that glossary terms come from the user's own project.

[0.12.5] - 2026-02-26

Changed

README — Add direct link to How context works docs page.
Docs site — Footer copyright year 2026; sitemap and docs README include How Context Works page.

[0.12.4] - 2026-02-26

Changed

Docs — Clarify that install is project-level only (README, CLI description, docs site quick-start and commands). No global mode; standards live in git with the project.
Docs site — New Squad Workflow page (/docs/squad) with pipeline roles, single-task and batch flows, and feedback loop. Slash-commands page now links to it instead of listing all squad commands inline.
Docs site — New How Context Works page (/docs/how-context-works) explaining the two-layer architecture (bridge file + scoped rules).

Added

Install — Warn when running ck install with no .git or package.json in the current directory (reminder to run from inside a project).

[0.12.3] - 2026-02-26

Fixed

Tests — Unknown-command CLI test now accepts "ContextKit not initialized" when not installed (CI), so it passes in both CI and local.

[0.12.2] - 2026-02-26

Fixed

CLI — Register opencode command so ck opencode works (integration existed; command was documented but not registered).
Tests — CLI integration tests for status and unknown-command now pass in both CI (no config) and local (full install).

[0.12.1] - 2026-02-23

Fixed

Pre-push quality gates — TypeScript, ESLint, and Prettier gates now only run when listed as dependencies in package.json, not when they appear in keywords. Fixes false positives for projects that mention these tools in keywords but don't use them.

[0.12.0] - 2026-02-23

Added

OpenCode integration — New platform support. Run ck opencode to create AGENTS.md (auto-loaded by OpenCode). Same bridge file pattern as Codex.

[0.11.0] - 2026-02-23

Added

/squad-run-agents — New Claude Code-only command that runs the squad batch pipeline using parallel subagents. Spawns one architect agent per task simultaneously (Phase 1), then one full dev→test→review pipeline agent per task simultaneously (Phase 2). Significantly faster than /squad-run for multi-task batches.
ESLint accessibility linting for contextkit-docs — installed eslint-plugin-test-a11y-js with the flat/react preset and @typescript-eslint/parser. Fixed 30 violations across 11 files (heading order, table structure, button/form/link labels, keyboard accessibility).

Fixed

.gitignore docs artifact paths corrected from vibe-kit-docs/ → contextkit-docs/

[0.10.0] - 2026-02-20

Changed

Interactive platform picker — ck install now prompts "Which AI tool?" instead of auto-detecting all tools. Supports a positional argument (ck install claude) to skip the prompt. --non-interactive installs base files only. Platform-specific commands (ck claude, etc.) are unchanged.

[0.9.7] - 2026-02-19

Added

/ck Health Check — slash command to check project health from inside any AI tool
- Detects whether ContextKit is installed
- Checks if standards files are still skeletons (wasting context tokens)
- Checks if product files still have placeholder content
- Lists active platform integrations
- Reports a status table with recommended next steps
Universal command file .contextkit/commands/health-check.md created during install
Platform wrappers: .claude/commands/ck.md and .cursor/prompts/ck.md

[0.9.6] - 2026-02-18

Changed

Claude @imports — CLAUDE.md now uses @path syntax to auto-load standards into context
- Standards, product files, and corrections.md are imported automatically every session
- Eliminates manual Read tool calls — saves tokens per session by avoiding repeated file reads
- First-time use triggers a one-time approval dialog in Claude Code
- .claude/rules/ files simplified to reference auto-loaded standards instead of duplicating paths
- Base integration (getStandardsBlock()) unchanged — @ imports are Claude-specific

Token Impact

Upfront: slightly higher base context (standards content loaded immediately)
Per-session: lower total usage — no Read tool calls needed for standards files
Net effect: fewer tokens overall for typical sessions, plus guaranteed consistency

[0.9.5] - 2026-02-18

Added

Squad Workflow — multi-agent pipeline with 7 slash commands
- /squad — kick off a task as Product Owner (writes PO spec)
- /squad-architect — design the technical plan from the PO spec
- /squad-dev — implement code following the architect plan
- /squad-test — write and run tests against acceptance criteria
- /squad-review — review the full pipeline and give a PASS/NEEDS-WORK verdict
- /squad-batch — kick off multiple tasks at once (batch PO specs)
- /squad-run — auto-run the remaining pipeline for batch tasks
Role-to-role feedback loop: downstream roles can raise questions for upstream roles via *-clarify statuses
Shared handoff file (.contextkit/squad/handoff.md) tracks specs, plans, implementation, tests, and review

Docs

Added squad commands to README slash commands table and new Squad Workflow section
Added squad commands and Squad Workflow section to docs site slash commands page

[0.9.4] - 2026-02-16

Added

Slash Commands for Claude Code (.claude/commands/) and Cursor (.cursor/prompts/)
- /analyze — scan codebase and generate standards content
- /review — code review with checklist
- /fix — diagnose and fix bugs
- /refactor — refactor code with safety checks
- /test — generate comprehensive tests
- /doc — add documentation
Both platforms delegate to universal .contextkit/commands/ files
New command files: review.md, fix.md, refactor.md

Changed

Git hooks now use git config core.hooksPath .contextkit/hooks instead of writing to .git/hooks/
Auto-injects prepare script into package.json so hooks work for all devs after npm install
Hook files renamed from .sh extension to match git conventions (pre-push, commit-msg)
All existing commands rewritten to be framework-agnostic (no React assumptions)
Legacy .git/hooks/ ContextKit wrappers automatically cleaned up

Fixed

Gradle quality gate now checks for gradlew and verifies check task exists
Go and Maven gates skip when no source files exist

[0.9.0] - 2026-02-16

Added

Quality Gates: Pre-push hooks now auto-detect your project framework and run the right checks automatically
- Node.js: TypeScript, ESLint, Prettier, build, test (auto-detects npm/yarn/pnpm/bun)
- Python: ruff/flake8, mypy, black/ruff format, pytest
- Rust: cargo check, clippy, cargo test
- Go: go vet, golangci-lint, go test
- PHP: PHPStan, PHPUnit
- Ruby: RuboCop, RSpec/rake test
- Java: Maven verify / Gradle check
- Generic: informational message for unrecognized projects
All gates skip gracefully when tools aren't installed
Integration test suite for framework detection and hook installation

Changed

Git hooks now use native approach instead of Husky
Hooks work in any git repo — no longer requires package.json or Node.js
ck install hooks prompt now checks for .git/ instead of package.json
Install prompt updated with "Quality Gates" branding

Removed

Husky dependency — no longer installed or required
installHusky(), checkHuskyInstalled(), initializeHusky(), checkCommandExists() methods from GitHooksManager

Fixed

Pre-push script grep patterns now match script keys ("test":) instead of any occurrence of the word in package.json

Migration

Existing .husky/ directories with ContextKit markers are automatically cleaned up on next ck install
Users can manually run npm uninstall husky to remove the dependency

FilesExpand file tree

CHANGELOG.md

Latest commit

History

CHANGELOG.md

File metadata and controls

Changelog

[1.7.0] - 2026-08-14

Added

Removed

[1.6.1] - 2026-08-11

Fixed

[1.6.0] - 2026-08-11

Changed

[1.5.0] - 2026-08-10

Changed

Fixed

[1.4.0] - 2026-08-05

Changed

[1.3.0] - 2026-07-24

Removed

[1.2.3] - 2026-07-22

Added

Changed

[1.2.2] - 2026-07-21

Changed

Fixed

[1.2.1] - 2026-07-14

Fixed

Added

Docs

Tests

[1.2.0] - 2026-07-13

Added

Tests

[1.1.7] - 2026-07-09

Changed

Added

[1.1.6] - 2026-06-29

Added

[1.1.5] - 2026-06-29

Fixed

Tests

[1.1.4] - 2026-06-29

Fixed

[1.1.3] - 2026-06-29

Added

Changed

Improved

[1.1.2] - 2026-06-25

Removed

[1.1.1] - 2026-06-24

Changed

[1.1.0] - 2026-06-24

Changed

Added

[1.0.3] - 2026-06-24

Changed

[1.0.2] - 2026-06-20

Changed

[1.0.1] - 2026-06-08

Added

[1.0.0] - 2026-06-08

Removed

Fixed

[0.21.0] - 2026-06-08

Security

[0.20.0] - 2026-05-26

Removed

Changed

[0.19.0] - 2026-05-17

Added

Changed

[0.18.0] - 2026-05-17

Added

[0.17.0] - 2026-05-17

Added

[0.16.4] - 2026-04-10

Fixed

[0.16.3] - 2026-04-10