Skip to content

Releases: alirezarezvani/claude-skills

Release list

v 2.9.0

v 2.9.0 Latest
Latest

Choose a tag to compare

View all tags
@alirezarezvani alirezarezvani released this 28 May 21:22
v2.9.0
0d78fd8
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

What's Changed

New Contributors

Full Changelog: v2.8.2...v2.9.0

Contributors

mitnick2012, alirezarezvani, and sandeepyadav1478
Assets 2
Loading

v2.8.2 — productivity/handoff skill, Matt Pocock-inspired

v2.8.2 — productivity/handoff skill, Matt Pocock-inspired

Choose a tag to compare

View all tags
@github-actions github-actions released this 23 May 01:48
v2.8.2
bd3d060
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

2.8.2 — productivity/handoff skill, Matt Pocock-inspired

Released 2026-05-23

Single-skill point release on top of v2.8.1. New productivity/handoff/ skill is a sibling to the existing engineering/handoff/ (shipped in v2.6.0). Both preserve Matt Pocock's seven-sentence body verbatim; the productivity variant adds the wrappers the engineering port deliberately skipped.

Added — productivity/handoff/ skill (PR #724)

Single-skill plugin. Stdlib-only across the board. 23 files total.

  • SKILL.md — Matt Pocock's seven-sentence body preserved verbatim, surrounded by invocation triggers, output-path discipline, 5-section template (Goal / State / Decisions / Skills / Artifacts), redaction checklist, anti-patterns block in Matt's register, examples, usage table.
  • scripts/setup.py — first-run Q&A. 5 core questions (save location, retention, redaction strictness, git context, recommender scope) plus 2 optional (filename style, project override). No pre-selected default for Q1: user explicitly picks OS temp / home folder / per-project .handoff/ / custom on first run. Prompt-once-then-default model: declining setup drops a sentinel so the prompt never re-appears. /cs:handoff-setup re-runs anytime.
  • scripts/handoff_template_generator.py — writes the 5-section scaffold at the configured path. Auto-includes git context (branch, last commit, dirty-file count) when in a repo + enabled in config.
  • scripts/redaction_linter.py — 17 stdlib regex patterns: AWS access keys, AWS secret assignments, GitHub tokens, OpenAI keys, Anthropic keys, Slack tokens, Google API keys, Stripe keys, private-key blocks, JWT, env-style secret assignments, DB connection strings with creds, bearer tokens, URL token params, email, phone, private CIDR. Strict-by-default with inline <!-- handoff:allow secret --> whitelist marker. Operationalizes Matt's redaction sentence (which the engineering port left as prose-only).
  • scripts/skill_recommender.py — scans repo SKILL.md files, scores against goal text, returns top 3-5 matches with one-line why. Hard cap at 5 — refuses to list more.
  • scripts/cleanup.py — mtime-guarded retention cleanup. Never deletes a handoff the user edited as a working surface (data-loss prevention).
  • scripts/config_loader.py — shared helper. Project config → global config → built-in defaults precedence.
  • hooks/session_start.py + hooks/hooks.json — SessionStart hook surfaces latest handoff (within retention window) as <handoff_from_previous_session> data. Treated as data, not instructions. Disable via HANDOFF_SESSIONSTART=0.
  • references/handoff_prompt.md — mandatory 7-step checklist for the agent. Forces topic-by-topic classification (State / Decision / drop) instead of free-handing prose.
  • references/handoff_structure.md — 5-section template with worked example.
  • references/deduplication_discipline.md — Matt's no-duplication rule made concrete via do-this-not-that pairs.
  • references/redaction_checklist.md — what regex catches + manual-review steps for what regex can't.
  • references/configuration.md — field-by-field config reference.
  • agents/cs-handoff-author.md + commands/cs-handoff.md + commands/cs-handoff-setup.md.

Added — v1.1 improvements (PR #728)

Three follow-up improvements judged most impactful in v1.1 design review.

  • hooks/session_end.py — pairs with SessionStart. When a session ends with no handoff in the last 30 minutes, prints a one-line reminder. Cannot prompt interactively or block session end (Claude Code hook constraint) — surfaces text via stdout. Disable via HANDOFF_SESSIONEND=0. hooks/hooks.json updated to wire both hooks.
  • scripts/handoff_self_check.py — operationalizes handoff_prompt.md. 6 checks: all 5 sections present, Goal non-empty, State bullets reference an artifact (commit hash / PR / file path), Open Decisions present when git is dirty / has recent commits, Skills 3-5 with — why explanation, Artifacts are paths/URLs only. Strict mode exits 1 only on high-severity findings (medium issues warn without blocking). Closes the fidelity gap (W3 from the design weakness list).
  • --refresh flag on handoff_template_generator.py — reuses the most recent handoff instead of creating a new file. Falls through to create-if-missing when none exists. Keeps the save location uncluttered.
  • /cs:handoff command flow updated to insert self-check between scaffold-fill and redaction linter.

Coexists with — engineering/handoff/

Aspect productivity/handoff/ (v2.8.2) engineering/handoff/ (v2.6.0)
Primary audience End-of-day / cross-machine session handoff Code/PR handoff
First-run setup Yes (5 questions, configurable save location) No (fixed mktemp)
Redaction enforcement Yes (linter + whitelist + strict/warn/off) No
SessionStart auto-load Yes (hook + retention-aware) No
SessionEnd reminder Yes (when no recent handoff) No
Mandatory checklist Yes (handoff_prompt.md, 7 steps) No
Self-check fidelity script Yes No
Retention cleanup Yes (mtime-guarded) No

Both stay. Cross-referenced in their READMEs.

Inspired by

Matt Pocock's handoff skill (MIT). Matt's seven-sentence body of SKILL.md is preserved verbatim. The wrapper around it (first-run setup, redaction enforcement, hooks, self-check, --refresh, retention) is original work in this repo.

Audit results

Plugin audit ran twice (after each PR) — final state:

  • Phase 2 (structure): 86.0/100 (GOOD) — above 75 threshold
  • Phase 3 (quality): 63.0/100 (C) — above 60 floor (sibling capture scores 46.4 on the same scorer)
  • Phase 4 (scripts): PASS — 7 scripts, 3 PASS + 4 PARTIAL (sibling config_loader import flagged as external — false positive)
  • Phase 5 (security): PASS — 0 critical, 0 high
  • Phase 6 (marketplace): PASS — plugin.json valid, marketplace entry at 2.8.2
  • Phase 7 (ecosystem): PASS — Codex + Gemini indexed, 0 broken internal links
  • Phase 8 (code review): PASS — all 7 scripts and 2 hooks work end-to-end

Documentation sync (PR #729)

  • CLAUDE.md Current Scope: 328 → 329 skills, v2.8.2 Highlights section added, footer bumped.
  • mkdocs.yml nav: productivity/handoff entry added under Productivity.
  • docs/skills/productivity/handoff.md generated.
  • README.md badges: Skills 313 → 329, Agents 46+ → 49+, Commands 60+ → 79+. Productivity table row updated to 5 skills (incl. handoff).
  • docs/index.md + docs/getting-started.md: counts and version references updated.
  • 0 remaining stale v2.7.5 references (an earlier version-number typo that this release also corrects).

Stats

  • 328 → 329 skills (productivity: +1)
  • 14 domains unchanged
  • Plugins in marketplace: 59 → 60
  • New artifacts: 23 files in productivity/handoff/

PRs

  • #724 (v1.0 — 5 must-haves)
  • #728 (v1.1 — SessionEnd + self-check + --refresh)
  • #729 (docs sync)
Loading

v2.8.1 — Engineering role-skill upgrade: karpathy-coder + Matt Pocock applied to fullstack / frontend / backend

v2.8.1 — Engineering role-skill upgrade: karpathy-coder + Matt Pocock applied to fullstack / frontend / backend

Choose a tag to compare

View all tags
@github-actions github-actions released this 21 May 13:31
v2.8.1
f398e16
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

2.8.1 — Engineering role-skill upgrade: karpathy-coder + Matt Pocock applied to fullstack / frontend / backend

Released 2026-05-20

Audited and upgraded

The three role-based engineering skills — senior-fullstack, senior-frontend, senior-backend — were audited against the karpathy-coder + Matt Pocock canon already shipping in this repo (engineering/karpathy-coder, engineering/grill-me, engineering/grill-with-docs, v2.8.0 BizOps/Commercial pattern). Three findings drove the upgrade:

  1. Generic role descriptions, not opinionated workflows. Existing SKILL.md files described capabilities; they did not enforce assumptions, success criteria, or kill criteria.
  2. No customization surface. A 4-person SaaS startup and a 200-person enterprise read the same recommendations.
  3. No invocation contract. Other agents/skills could not orchestrate fullstack / frontend / backend lenses via a typed surface.

Added — per-skill artifacts (21 new files: 7 × 3 skills)

For each of senior-fullstack, senior-frontend, senior-backend:

  • scripts/<role>_decision_engine.py — stdlib-only deterministic profile picker. Refuses to recommend without the four core assumptions (Karpathy #1). Surfaces kill criteria. Names the human approver chain (never auto-approves).
  • profiles/*.json × 4 — customization profiles, JSON-loadable, swappable. Users copy one to <your-org>.json to override defaults.
  • references/forcing_questions.md — 7 Matt Pocock forcing questions per skill, one per turn, each with recommended answer + canon citation + kill criterion. 21 forcing questions total across the three roles.
  • references/composition_map.md — explicit routing table to POWERFUL-tier specialists (api-design-reviewer, database-designer, slo-architect, performance-profiler, a11y-audit, epic-design, apple-hig-expert, etc.).

Added — 3 orchestrator agents (cs-* with context: fork)

  • agents/engineering/cs-fullstack-engineer.md — walks 7 fullstack questions → decision engine → forks specialists.
  • agents/engineering/cs-frontend-engineer.md — frontend equivalent.
  • agents/engineering/cs-backend-engineer.md — backend equivalent.

All three are invokable by other agents via Agent({subagent_type:"cs-<role>-engineer", prompt:"..."}) — the "invokable by other agents" promise.

Added — 4 slash commands

  • /cs:fullstack-review <prompt> — full grill + decision engine + composition routing.
  • /cs:frontend-review <prompt> — frontend equivalent.
  • /cs:backend-review <prompt> — backend equivalent.
  • /cs:engineer-grill <plan> [--lane fullstack|frontend|backend|all] — cross-role 21-question forcing-question runner.

Augmented — 3 existing SKILL.md files (additive edits only)

Each augmented SKILL.md now has 4 new sections appended:

  1. Assumptions and Verifiable Success Criteria (Karpathy #1 + #4) — names the four core assumptions + three machine-checkable success criteria.
  2. Customization profiles — table of the 4 profiles + how to add an org-specific one.
  3. Composition map — table of which POWERFUL specialist to fork into per sub-concern.
  4. Forcing-question library (Matt Pocock grill) — summary of the 7 questions + the discipline.
  5. Invocation from other agents and skills — explicit contract: 3 surfaces.

Principles enforced

  • Karpathy #1 (Think Before Coding): every decision engine refuses to run without the four core assumption inputs.
  • Karpathy #2 (Simplicity First): profiles never auto-recommend microservices unless team size + platform team + bounded-context independence all pass (Newman's MonolithFirst).
  • Karpathy #3 (Surgical Changes): the upgrade did NOT rewrite existing SKILL.md content. New sections appended; existing tools / references / scaffolding untouched.
  • Karpathy #4 (Goal-Driven Execution): every recommendation prints verifiable success criteria (latency floor, CWV target, SLO).
  • Matt Pocock grill discipline: all 21 forcing questions ship with recommended answer + canon citation + kill criterion + one-per-turn rule.

Verification

  • 12/12 profile JSON files parse cleanly (json.load).
  • 3/3 new Python decision engines pass --help and --sample, exit code 0.
  • 3/3 new cs-* agents have valid YAML frontmatter with required keys (name, description, skills, domain, model, tools, context: fork).
  • 3/3 agent → skill relative paths resolve from agents/engineering/.
  • 3/3 slash commands reference their correct cs-* agent.
  • Existing SKILL.md content unchanged (additive edits only — Karpathy #3, surgical scope).

Customization story

Adding org-specific defaults requires zero code changes:

cp engineering-team/skills/senior-fullstack/profiles/saas-startup.json \
   engineering-team/skills/senior-fullstack/profiles/my-org.json
# Edit constraints + stack_recommendations + named_approver_chain
# Decision engine auto-discovers the new profile via Path.glob('*.json')

This is the "world-class customizable plugin" promise: profiles are the customization surface, not code.

Versions bumped

  • engineering-team/.claude-plugin/plugin.json: 2.2.3 → 2.8.1
  • .claude-plugin/marketplace.json: engineering-skills plugin → 2.8.1
Loading

v2.8.0 — business-operations + commercial domains, plugin.json regression fix, auto-release pipeline

v2.8.0 — business-operations + commercial domains, plugin.json regression fix, auto-release pipeline

Choose a tag to compare

View all tags
@github-actions github-actions released this 19 May 04:05
v2.8.0
1816605
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

2.8.0 — business-operations + commercial domains, plugin.json regression fix, auto-release pipeline

Released 2026-05-19

Added

business-operations/ — new top-level domain (Sprint 1)

Internal-ops skills for BizOps leads, COO direct reports, vendor management, IT ops. Sprint 1 ships the orchestrator + 2 sub-skills using context: fork to chain without polluting parent context. Sprint 2 will add capacity-planner, internal-comms, knowledge-ops, procurement-optimizer.

  • business-operations-skills (orchestrator) — routes inquiries to the right sub-skill and returns a digest
  • process-mapper — BPMN modeling + bottleneck detection + cycle-time analysis
  • vendor-management — SLA tracking + risk scoring + supplier scorecards
  • Distinct from business-growth/ (external sales) and c-level-advisor/ (strategic, not operational)

commercial/ — new top-level domain (Sprint 1)

Per-deal economics skills for pricing, deal desk, partnerships. Sprint 1 ships the orchestrator + 2 sub-skills. Sprint 2 will add partnerships-architect, channel-economics, commercial-policy, rfp-responder, commercial-forecaster.

  • commercial-skills (orchestrator) — routes commercial inquiries via context: fork
  • pricing-strategist — Van Westendorp WTP analysis + packaging + pricing-model picker
  • deal-desk — margin analysis + discount routing + contract redline scoring
  • Distinct from business-growth/sales-engineer, c-level-advisor/cro-advisor, finance/financial-analysis

Forcing-question slash commands

  • /cs:grill-bizops — Matt Pocock docs-anchored grilling for BizOps workflows
  • /cs:grill-commercial — same for commercial decisions (pricing, deals, partnerships)

Release automation

  • .github/workflows/release.yml — On every push to main, parses CHANGELOG.md and auto-creates a git tag + GitHub Release for the latest version listed. Idempotent (skips if tag already exists). Release notes are extracted from the matching CHANGELOG section.
  • scripts/extract_release_notes.py — Stdlib-only CHANGELOG parser. Extracts the latest version, date, subtitle, and body. Used by the release workflow but also runnable standalone for previewing release notes.

Fixed

Plugin manifest /doctor warning — issue #686 (reported by @esoneill)

Claude Code 2.1.133+ rejects "skills": "./skills" with a "Path escapes plugin directory" warning, even though ./skills resolves to a valid subdirectory inside the plugin root. This blocked skill registration for the 9 main marketplace plugins plus 38 sibling sub-plugins.

  • PR #689 — replaced "skills": "./skills" with "skills": "skills" across all 47 affected plugin.json files. Updated CLAUDE.md ClawHub publishing constraints to document the new convention.
  • PR #690 (regression prevention) — scripts/check_plugin_json.py now actively rejects any "skills" string starting with "./" (catches both "./skills" and "./skills/sub" regressions). Wired into ci-quality-gate.yml as a blocking step on every PR. Also recognized source and attribution as approved extension fields (per CLAUDE.md), and dropped the over-strict "./" rejection inside arrays (["./"] is the documented single-skill-at-root form). Validator's previous error message was actually recommending "./skills" verbatim — a leftover from #539, the first round of this same upstream rule tightening — which has been corrected.

This is the second round of the same Claude Code path-validator tightening (round 1 was #539, fixing "./" → "./skills" at CC v2.1.107). The new validator + CI gate prevents a future round 3 from silently shipping again.

Maintenance

  • inspect-assets.py (#684, contributor: @TemaDeveloper) — --help now works without Pillow installed
  • Codex symlink syncs (automated)

Stats

  • 313 → 319 skills (business-operations: +3, commercial: +3)
  • 12 → 14 top-level domains
  • 60 → 68 slash commands

Contributors

esoneill and TemaDeveloper
Loading

v2.0.0 — 86 Skills, 9 Domains, 26 POWERFUL-Tier

v2.0.0 — 86 Skills, 9 Domains, 26 POWERFUL-Tier

Choose a tag to compare

View all tags
@alirezarezvani alirezarezvani released this 04 Mar 02:22
v2.0.0
3960661
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

🚀 Claude Skills Library v2.0.0

The biggest release since launch — more than doubling the skill count with deep engineering capabilities.

📊 By the Numbers

Metric v1.0.2 v2.0.0
Skills 42 86
Domains 6 9
Python tools 20+ 92+
POWERFUL-tier 0 26
Refactored 0/42 37/42

⚡ 26 POWERFUL-Tier Skills (NEW)

Advanced engineering skills with comprehensive tooling:

  • agent-designer — Multi-agent architect & performance evaluator
  • rag-architect — RAG pipeline builder & retrieval optimizer
  • database-designer — Schema analyzer with ERD generation
  • migration-architect — Migration planner & rollback generator
  • skill-security-auditor — 🔒 Security gate for skill installation (NEW)
  • ci-cd-pipeline-builder — Stack-aware pipeline generator
  • mcp-server-builder — MCP server scaffolder
  • pr-review-expert — Structured code review system
  • And 18 more...

🆕 3 New Domains

  • business-growth — Customer success, sales engineering, revenue ops
  • finance — Financial analyst with DCF valuation
  • engineering (POWERFUL) — 26 advanced engineering skills

🔄 Anthropic Best Practices Refactoring

37 of 42 original skills rewritten with:

  • Progressive Disclosure Architecture
  • Trigger phrases for better activation
  • Numbered workflows with validation checkpoints
  • Concise SKILL.md (<200 lines) with reference files

🤖 Platform Support

  • OpenAI Codex — Full compatibility with .codex/skills/ symlinks
  • Claude Code — Native marketplace plugin support
  • OpenClaw — Installation script included

📦 Install

# Claude Code
/plugin marketplace add alirezarezvani/claude-skills

# OpenClaw
bash <(curl -s https://raw.githubusercontent.com/alirezarezvani/claude-skills/main/scripts/openclaw-install.sh)

# Codex
npx agent-skills-cli add alirezarezvani/claude-skills --agent codex

Full changelog: CHANGELOG.md

Loading