- Migrate all 20 command files from
commands/intoskills/— completes the command-to-skill conversion started in v2.0.0 - Remove
"commands"key frommanifest.json— all entries now under"skills" - Fix
manifest.jsonversion (was stuck at1.3.5whileplugin.jsonwas at2.0.0) - Remove duplicate
skills/block in README repository structure - Fix orphaned
skill-authoringreference in CONTRIBUTING.md
- Add
devkit:audit— unified project health audit (deps, vulnerabilities, outdated packages, licenses, lint, security) with scored report - Add
devkit:pr-monitor— post-PR review monitor that watches CI and iteratively resolves reviewer comments - Add
devkit:repo-map— AST-based symbol index with dependency graph, cached to .devkit/repo-map.json
- Add
security-patterns— PreToolUse on Edit/Write catches eval, XSS, shell injection, weak hashes, hardcoded secrets across JS/TS/Python/Go - Add
audit-trail— logs all Bash commands to.devkit/audit.logwith UTC timestamps, auto-rotates at 10k lines - Add
slop-detect— PostToolUse on Edit/Write catches excessive docs, restating comments, JSDoc overuse - Add
pr-gate— prompts to run pr-ready pipeline beforegh pr create, 10-minute cooldown
- Upgrade
self:perfto hypothesis-driven investigation — evidence gathering, ranked hypotheses, one-at-a-time testing with 3x benchmark runs
- Define holistic setup: devkit + superpowers + feature-dev + pr-review-toolkit + commit-commands + hookify + skill-creator + context-mode
- Evaluated all official marketplace plugins — documented which to install, which to skip, and why
- Fix superpowers install: use
@claude-plugins-official, not separate marketplace
- Convert 6 commands to context-activated skills (no slash command needed): test-gen, doc-gen, changelog, onboard, research, scrape
- Add
dont-reinventskill — prefer existing solutions over custom code, reduce maintenance burden - Add
gcliskill — Google Workspace CLI reference with safety confirmation gate - Remove 5 skills that overlap with superpowers: brainstorming, planning, writing-tests, skill-authoring, verify
- Evaluated
code-simplifieras replacement — rejected (thin, React-specific, no test verification) - Commands reduced from 26 to 20, skills increased from 6 to 14
- Fix stop-gate: disabled — fires every turn, not just session end. Needs architectural redesign.
- Fix stop-gate conflict marker false positive — grep pattern was matching its own source code
- Complete README overhaul with companion ecosystem diagram and holistic setup guide
- Added "why these and not others" section explaining plugin selection rationale
- Updated roadmap with completed and planned items
- Add Go CLI harness: deterministic loop control, multi-agent dispatch, SQLite state
- Add test suite for runners and lib packages with race detection
- Add CI pipeline (GitHub Actions): build, vet, test, gofmt check
- Add
test-gencommand as composable Go primitive - Add
feature,bugfix,refactorcommands as composable Go primitives - Fix security findings: budget enforcement, permissions, DB error handling
- Fix consensus findings from tri-review: races, exit codes, verify-before-commit
- Add Gemini agent integration alongside Codex
- Add RTK token optimization hook — 60-90% savings on Bash output (optional, no-op if rtk not installed)
- Add 12 YAML workflow definitions from pikit: feature, bugfix, refactor, research, self-improve, self-test, self-lint, self-perf, tri-review, tri-dispatch, tri-debug, tri-security
- Move marketplace to separate repo (
5uck1ess/marketplace) for multi-plugin support - Update install instructions to use new marketplace URL
- Add RTK to
/devkit:statushealth check
- Add
decomposecommand — goal decomposition into task DAG with dependency ordering - Add 4 new skills:
skill-authoring,creating-workflows,stuck,verify - Add
ROADMAP.mdandPREFERENCES.md - Add Budget & Early Exit sections to all self:* commands with stuck detection
- Add Concurrency & Budget sections to all tri:* commands with
[PARALLEL]markers - Add token budget guidance to bugfix, feature, refactor, research commands
- Update research command with structured
AskUserQuestioncalls
- 21 commands: solo workflows, self-improvement loops, multi-agent dispatch
- 7 skills: planning, executing, writing-tests, clean-code, dry, yagni, brainstorming
- 6 agents: reviewer, researcher, improver, test-writer, documenter, security-auditor
- Safety hooks: PreToolUse blocking destructive operations
- Graceful degradation: tri:* commands work with 1-3 agents
- Codex integration via official codex-plugin-cc
- Workflow runner for custom YAML workflows