Version: 0.116.1
Complete DevOps automation plugin for Claude Code. Hooks, skills, agents, and templates that make shipping faster, safer, and smarter.
⚠ AI runs commands on your machine. Hooks, skills, and agents execute shell commands, edit files, push to remotes, and launch apps autonomously. Built-in safeguards reduce risk but do not replace your review.
Work in a versioned tree. Keep backups. Read what Claude proposes before approving. Use at your own risk.
Token math — costs and payoff:
- Costs: ~0.7M tokens/week — hooks, prompt guards, self-calibration
- Plan share: 1–4% on Max plans · up to ~23% on Pro + Opus
- Saves (context): ~1–4M tokens/week — token guard blocks expensive reads before they land
- Saves (time): 2–4 hours/week — auto git state, ship pipeline, root-cause debug
- Net: plugin pays itself back 1.5–6× via prevented context waste
Detailed token math — weekly breakdown, plan %, what you get back
| Source | Tokens/week | Notes |
|---|---|---|
| Startup hooks (4x per session) | ~8K | Update check, git check, token scan, MCP deps |
| Prompt guards (per message) | ~150K–250K | Ship detection, issue tracking, git sync — most exit silently |
| Tool guards (per tool call) | ~100K–200K | Token budget + ship enforcement — early-exit when clean |
| Self-calibration (every 10 min) | ~200K–400K | Deep-knowledge rotation, skill internalization |
| Skill invocations (~15–25/week) | ~15K–30K | Only when you call /devops-ship, /devops-commit, etc. |
| Total | ~500K–900K | ~0.7M tokens/week on average |
Based on ~0.7M tokens/week plugin overhead:
| Plan | Model | Weekly budget (approx.) | Plugin overhead |
|---|---|---|---|
| Pro ($20) | Sonnet | ~10M | ~7% |
| Pro ($20) | Opus | ~3M | ~23% |
| Max 5x ($100) | Sonnet | ~55M | ~1.3% |
| Max 5x ($100) | Opus | ~18M | ~3.9% |
| Max 20x ($200) | Sonnet | ~225M | ~0.3% |
| Max 20x ($200) | Opus | ~75M | ~0.9% |
Budgets are rough estimates and vary by usage pattern. Anthropic adjusts limits dynamically.
| Without plugin | With plugin |
|---|---|
| "Wait, did I push that?" | Git state checked on every session start |
git push --force to main at 2 AM |
Blocked before it happens |
| Forgetting to bump the version | /devops-ship handles version, PR, merge, cleanup |
| "Why is my context window gone?" | Token guard kills expensive reads before they land |
| Debugging the same error 4 times | /devops-flow kicks in after the second failure |
| Writing commit messages by hand | Conventional commits, auto-staged, one command |
Token guard payoff: The token guard blocks any single operation above your plan's per-operation share of the ~200K context window — 5% (~10K tokens) on Pro, 8% (~16K) on Max 5×, 10% (~20K) on Max 20×. In a typical session, Claude attempts 5–15 broad searches or large-file reads that would each burn 20–80K tokens — that's 100–400K tokens/session evaporating into context you never asked for. Across ~10 sessions/week, the guard saves roughly 1–4M tokens/week in prevented waste. The plugin's own overhead (~0.7M tokens/week for hooks, startup checks, and skill prompts) pays for itself 1.5–6x over just by keeping Claude from reading files it doesn't need.
Your mileage may vary. Your sanity will not.
Setup
Use
Details
Add the plugin via CLI (recommended):
claude plugin add devops@Jerry0022Desktop App: The marketplace UI shows the marketplace tab but may not list plugins for installation. Use the CLI command above, or see
INSTALL.mdfor manual registration steps.
Start a new session for hooks to take effect. See INSTALL.md for details, extensions, and uninstall.
claude plugin update devops@Jerry0022Or enable auto-update via Settings → Plugins → Marketplaces. Semantic versioning — breaking changes only in major versions.
This plugin is built and actively tested against a specific stack. Outside that stack, hooks and skills may work, degrade gracefully, or not run at all — anything not listed as supported is best-effort.
| Area | Supported | Behavior outside |
|---|---|---|
| OS | Windows, macOS, Linux | Other platforms: AnythingLLM lifecycle reports unsupported-platform; core git / skill flows still run via Node + git |
| Shell | bash (Git-Bash on Windows), zsh | PowerShell / cmd are untested — use Git-Bash or WSL on Windows |
| Git hosting | GitHub (via gh CLI) |
GitLab / Gitea / Bitbucket / self-hosted: issue tracking, PR creation, and ship release will fail. Local / push-only flows still work |
| Default branch | Auto-detected from origin/HEAD (main, master, or any other name) |
Detached HEAD or missing origin/HEAD: falls back to main |
| Build system | npm — auto-detects build, lint, test scripts in package.json |
No package.json: build / lint / test steps are silently skipped. pnpm, yarn, pytest, cargo, go test, maven, gradle etc. are not invoked |
| Local LLM | AnythingLLM Desktop (HTTP API) | Feature degrades gracefully — local_generate becomes unavailable, all main flows continue normally |
| Node runtime | Node.js 20+ | Older Node: MCP server and hooks may fail to start |
If your stack differs, extend the plugin per-project via the 3-layer extension model — see Customization.
Install codex-plugin-cc alongside this plugin for AI-powered code review and task delegation via OpenAI Codex. Both plugins coexist as independent skill providers — no configuration needed.
Combined workflows: /codex:rescue for pre-ship code review, parallel investigation
for research tasks, and QA-integrated review for complex changes.
See INSTALL.md for setup instructions.
Every skill and agent supports three-layer extensions:
Layer 1: Plugin (this plugin) ← defaults
Layer 2: User global (~/.claude/) ← your personal overrides
Layer 3: Project ({project}/.claude/) ← project-specific rules
To extend any plugin skill for your project, create a directory matching the
skill's name under .claude/skills/ in your project:
your-project/.claude/skills/{skill-name}/
├── SKILL.md ← override or add steps
└── reference.md ← project-specific context
The plugin reads your extensions before executing and merges them. Your rules win on conflict. Both files are optional — create only what you need.
Example — extending /devops-ship with project-specific quality gates and deploy targets:
your-project/.claude/skills/devops-ship/
├── SKILL.md ← "Before PR: run ng build --prod"
└── reference.md ← "Deploy via SSH to 192.168.178.32"
Run /devops-extend-skill to interactively scaffold an extension for any plugin skill.
It detects existing extensions and lets you adapt them.
For the full extension guide with examples per skill, see deep-knowledge/skill-extension-guide.md.
- 35 Hooks — automated guards and triggers across the full session lifecycle
- 22 Skills — devops-ship, devops-commit, devops-flow, devops-new-issue, devops-project-setup, devops-readme, devops-refresh-usage, devops-extend-skill, devops-repo-health, devops-claude-md-lint, devops-concept, devops-agents, devops-plugin-update, devops-autonomous, devops-burn, devops-learn, devops-harden, devops-polish, devops-test-plan
- 12 Agents — AI, Core, Designer, Feature, Frontend, Gamer, PO, QA, Redteam, Research, Windows
- Completion Flow — mandatory card after every task (8 variants), visual verification, ship recommendation
- Ship Enforcement — intent detection, PR command blocking, automatic /devops-ship skill routing
- 3-Layer Extension Model — customize any skill or agent per-project without forking
By session lifecycle — when does it fire?
SessionStart ──> UserPromptSubmit ──> PreToolUse ──> PostToolUse ──> Stop
ss.plugin.update— Auto-update plugin marketplace clones, rebuild cache, and update registry.ss.permissions.ensure— Ensure required plugin permissions exist so devops skills that write ephemeral review…ss.statusline.ensure— Enables the native usage source.ss.knowledge.index— Inject deep-knowledge INDEX.md into context at session start.ss.mcp.deps— Auto-install MCP server dependencies into CLAUDE_PLUGIN_DATA, and self-heal partial i…ss.mcp.envcheck— Detect enabled plugins whose .mcp.json references env vars that are not set.ss.mcp.verify— Verify every MCP server declared in this plugin's .mcp.json has its entry file presen…ss.tokens.scan— Scan project for expensive files and update config for the pre.tokens.guard hook.ss.git.check— Check for stale changes AND workspace setup issues at session start.ss.git.sync— Registers a recurring git sync cron job (every 10 minutes).ss.graphify— graphify enforcement — install-check + auto-build wiring for the devops-graph feature.ss.ship.verify— Surface results from the post-merge watcher (post-ship CI + optional deploy verify).ss.concept.resume— Recover an open concept session after a Claude restart.ss.team.changelog— Show a summary of changes made by other contributors on remote main since the last ti…
prompt.flow.silent-turn— Detects background/cron-injected prompts and marks the turn as "silent" so post.flow.…prompt.knowledge.dispatch— On-demand deep-knowledge injection based on prompt keywords.prompt.git.sync— Throttled git sync on user prompt — delegates to scripts/git-sync.js.prompt.issue.detect— Detect issue references in user messages.prompt.skill.enforce— Detects /devops-* skill commands mentioned INLINE in a user prompt (typed as text, no…prompt.ship.detect— Detect ship intent in user prompts and inject Skill('devops-ship') instruction.prompt.flow.appstart— Detect app start intent in user prompts.prompt.worktree.branch-guard— Prevents working without a dedicated branch inside a linked worktree.
pre.tokens.guard— Block Read/Bash/Glob/Grep operations that would consume a significant percentage of t…pre.ship.guard— Block manual PR creation/merging via Bash.pre.main.guard— Prevent accidental writes on local main/master.pre.worktree.split-guard— WARN (never block) on git-mutating work driven from the main repo root while an agent…pre.edit.branch— Prevent Edit/Write tool calls while HEAD is on local main/master.pre.mcp.health— Detects dead or stale MCP servers before tool calls fail cryptically.
post.flow.completion— After EVERY tool call: inject the completion-card reminder so Claude always has the i…post.flow.debug— After 2+ consecutive Bash failures: recommend the flow skill.post.graphify.query— When Claude runsgraphify query ..., record a per-session flag so the PreToolUse gr…post.concept.gate— Deterministic backstop for devops-concept pages.
stop.flow.browsertest— Light-verification enforcement gate (the "V" of the V&V gate).stop.flow.guard— Per-turn completion card + validation enforcement (the validation half of the V&V gate).stop.flow.selfcalibration— Run self-calibration when Claude finishes a response turn.
By category — what does it guard?
ss.tokens.scan— Scan project for expensive files (SessionStart)pre.tokens.guard— Block operations exceeding token budget (PreToolUse)
ss.git.check— Check for uncommitted/unpushed changes (SessionStart)ss.git.sync— Register recurring git-sync cron, every 10 min (SessionStart)prompt.git.sync— Throttled pull/merge main, every 15 min (UserPromptSubmit)
pre.ship.guard— Block manual PR/merge via Bash (PreToolUse)prompt.ship.detect— Detect ship intent, enforce /devops-ship skill (UserPromptSubmit)ss.ship.verify— Surface post-merge watcher results (SessionStart)
post.flow.completion— Track code edits, inject completion reminder (PostToolUse)post.flow.debug— Recommend /devops-flow after repeated failures (PostToolUse)prompt.flow.appstart— Detect app start intent, enforce completion card (UserPromptSubmit)prompt.flow.silent-turn— Mark background/cron-injected turns (UserPromptSubmit)stop.flow.guard— Enforce completion card before response ends (Stop)stop.flow.selfcalibration— Run self-calibration at end of turn (Stop)
pre.main.guard— Prevent accidental writes on local main (PreToolUse)pre.edit.branch— Block Edit/Write while HEAD is on main (PreToolUse)prompt.worktree.branch-guard— Prevent working on main inside a worktree (UserPromptSubmit)
ss.plugin.update— Check for plugin updates (SessionStart)ss.permissions.ensure— Ensure required plugin permissions (SessionStart)ss.mcp.deps— Auto-install MCP server dependencies (SessionStart)ss.mcp.envcheck— Warn on MCP servers missing env vars (SessionStart)pre.mcp.health— Detect dead/stale MCP servers before calls (PreToolUse)
ss.knowledge.index— Inject deep-knowledge index into context (SessionStart)prompt.knowledge.dispatch— Inject deep-knowledge by prompt keywords (UserPromptSubmit)
prompt.issue.detect— Track GitHub issues automatically (UserPromptSubmit)ss.concept.resume— Recover an open concept session after restart (SessionStart)ss.team.changelog— Summarize teammates' changes on remote main (SessionStart)
| Skill | Invocation | Purpose |
|---|---|---|
/devops-ship |
Explicit + Hook | Full shipping pipeline: build, version, PR, merge, cleanup |
/devops-release |
Explicit | Channel promotion (alpha→beta→stable): re-tag the same SHA, no rebuild |
/devops-commit |
Explicit | Conventional commits with smart staging |
/devops-flow (alias: /debug) |
Explicit + Hook | Root-cause analysis, diagnostics, and fix cycle |
/devops-new-issue |
Explicit | GitHub issue creation with labels and milestones |
/devops-project-setup |
Explicit | Repo hygiene audit and initialization |
/devops-readme |
Explicit | Modern README generation |
/devops-refresh-usage |
Explicit + Hook | Token usage tracking (CLI + CDP) |
/devops-extend-skill |
Explicit | Scaffold or adapt project-level skill extensions |
/devops-repo-health |
Explicit | Repository branch hygiene analysis and cleanup |
/devops-plugin-update |
Explicit | Update the plugin to the latest version from GitHub |
/devops-claude-md-lint |
Explicit | Audit CLAUDE.md files for size, structure, and token efficiency |
/devops-concept |
Explicit | Interactive HTML page for analysis, plans, concepts, and prototypes |
/devops-agents |
Explicit | Evaluate agents and orchestrate parallel execution |
/devops-autonomous |
Explicit | Fully autonomous agent orchestration while user is AFK |
/devops-burn |
Explicit | High-throughput autonomous task runner with aggressive parallelization |
/devops-learn |
Explicit | Capture long-term learnings and route to project-specific instructions |
/devops-harden |
Explicit | Stabilization pass: full test suite, autonomous bug fixes, regression + consistency |
/devops-polish |
Explicit | UI refinement: visual consistency, state-visuals, UI-side functionality checks |
/devops-test-plan |
Explicit + Hook | Detect test profile, deterministic tool-chain recommendations per test request |
/devops-graph |
Explicit + Hook | On-demand code knowledge graph via graphify, with opt-in auto-build + hard-gate enforcement |
/devops-rethink |
Explicit | Strategic reset for stuck development: code-blind fresh approaches, concept decision, autonomous implementation |
| Agent | Role |
|---|---|
| ai | AI/ML integration |
| core | Business logic and APIs |
| designer | UX/UI design, tokens, and specs |
| feature | Orchestrate feature implementation |
| frontend | UI components and styling |
| gamer | Player perspective and UX |
| po | Requirements and validation |
| qa | Test, verify, screenshot |
| redteam | Adversarial review: failure modes, blind spots, hidden risks |
| research | Deep-dive investigations |
| rethinker | Code-blind fresh-approach ideation through one lens |
| windows | Platform-specific features |
Every task ends with a completion card — a structured signal showing what happened, the repo state, and what comes next. The card is always the last thing in the response.
ship-successful — after a successful PR merge:
---
## ✨✨✨ Filter dialog moved to settings ✨✨✨
**Changes**
* Settings → Filter tab as new section with drag & drop
* Dialog → FilterDialog removed, route redirected to Settings
* Tests → 3 unit tests added for new settings section
**Tests**
* Build → npm run build successful
* Unit → 47/47 passed, 3 new tests
* Preview → Filter tab rendered correctly, drag & drop functional
✅ merged → origin/main · PR #42 "Filter dialog to settings" · a3f9b21 · feat/filter-settings
```
5h ━━━━━╏──────── 33% +2% · 3h 12m left
Wk ━━━─╏───────── 18% +2% · 5d 3h left
```
---
📌 0.8.2 → 0.8.3 (patch) · `a3f9b21`
## 🚀 SHIPPED. merged → origin/main — All DONE
---
See all other variants — ready, ship-blocked, test, test-minimal, analysis, aborted, fallback, ship-successful (direct push)
---
## ✨✨✨ Live search in contact book ✨✨✨
**Changes**
* ContactBook → Debounced live search with 300ms delay
* API → New /contacts/search endpoint with fuzzy match
* UI → Search field with clear button and loading spinner
**Tests**
* Build → successful
* Unit → 38/38 passed
🟢 not merged · no PR · not pushed · e82a0f7 · feat/contact-search · app running
**Please test**
1. Open contact book, type "Mue" in search field
2. Verify: results filter after ~300ms, spinner visible
3. Click clear button, verify list resets
```
5h ━━━━━━━╏────── 50% +3% · 2h 30m left
Wk ━━━━─╏──────── 27% +3% · 4d 16h left
```
---
📌 `e82a0f7`
## 🧪 DONE — SHIP after your TEST?
---
---
## ✨✨✨ Dark mode for dashboard implemented ✨✨✨
**Changes**
* Theme → Dark mode palette as CSS custom properties
* Dashboard → All panels switched to theme variables
* Settings → Dark mode toggle with localStorage persistence
**Tests**
* Build → successful
* Unit → 52/52 passed
🔀 not merged · no PR · not pushed · c91d3e8 · feat/dark-mode
```
5h ━━━━━━━━╏───── 58% +4% · 2h 5m left
Wk ━━━─╏───────── 22% +4% · 5d 1h left
```
---
📌 `c91d3e8`
## 📦 READY — SHIP or CHANGE?
---
---
## ✨✨✨ API rate limiting added ✨✨✨
**Changes**
* Middleware → RateLimiter with token bucket algorithm
* Config → Rate limits configurable per endpoint
* Tests → Integration tests for throttling behavior
**Tests**
* Build → successful
* Unit → 3/5 FAILED — testBurstLimit and testConcurrentRequests timeout
🔀 not merged · no PR · not pushed · d44f1a2 · feat/rate-limiting
```
5h ━━━━━━╏─────── 42% +3% · 2h 48m left
Wk ━━━─╏───────── 24% +3% · 4d 19h left
```
---
📌 `d44f1a2`
## ⛔ BLOCKED. 2 tests failed — FIX or SKIP?
---
---
## ✨✨✨ Push notifications for mobile ✨✨✨
**Changes**
* Mobile → Firebase Cloud Messaging integration
* Backend → Notification service with device token management
* Settings → Push notification opt-in/opt-out toggle
**Tests**
* Build → successful
* Unit → 29/29 passed
🟡 not merged · no PR · not pushed · f1b3d99 · feat/push-notifications · app not started
**Please test**
1. Start app on phone, grant push permission
2. Send a test message from another device
3. Verify: push appears, tap opens correct view
```
5h ━━━━━━━━━╏──── 67% +5% · 1h 40m left
Wk ━━━━─╏──────── 30% +5% · 4d 9h left
```
---
📌 `f1b3d99`
## 🧪 DONE — SHIP after your TEST?
---
---
## ✨✨✨ Dev server started ✨✨✨
📌 `a3f9b21`
## ▶️ STARTED. Website opens in Edge — HAVE FUN
---
---
## ✨✨✨ Hook lifecycle analyzed ✨✨✨
**Changes**
* Analysis → PostToolUse hooks fire only on Edit/Write, not Read
* Analysis → Session state shared via temp files, not env vars
* Analysis → prompt.ship.detect recognizes "ja"/"go" as ship intent
➖ No changes to repo
```
5h ━━━━╏───────── 25% +2% · 3h 45m left
Wk ━━━─╏───────── 19% +2% · 5d 6h left
```
---
📌 `a3f9b21`
## 📋 DONE — READ through
---
---
## ✨✨✨ WebSocket migration aborted ✨✨✨
**Changes**
* Analysis → existing REST polling architecture incompatible with WS
* Spike → prototype shows migration requires new session management
🔀 not merged · no PR · not pushed · abc1234 · spike/websocket
```
5h ━━━━━━━━━━━╏── 75% +6% · 1h 15m left
Wk ━━━━━─╏─────── 35% +6% · 4d 2h left
```
---
📌 `abc1234`
## 🚫 ABORTED. Architecture incompatibility — What should I TRY?
---
---
## ✨✨✨ Configuration verified ✨✨✨
**Changes**
* .editorconfig → settings match team standard
* .gitattributes → LF normalization correctly configured
➖ No changes to repo
```
5h ━━╏─────────── 15% +1% · 4h 10m left
Wk ━━──╏───────── 16% +1% · 5d 0h left
```
---
📌 `a3f9b21`
## 🔧 DONE — Anything ELSE?
---
---
## ✨✨✨ Typo in error handler fixed ✨✨✨
**Changes**
* ErrorHandler → "Unerwareter" corrected to "Unerwarteter"
✅ merged → origin/main · no PR · b7e2c44
```
5h ━╇──────────── 12% +1% · 4h 31m left
Wk ━━──╏───────── 15% +1% · 4d 22h left
```
---
📌 `b7e2c44`
## 🚀 SHIPPED. merged → origin/main — All DONE
---
8 variants total. The card always fires — see completion-card.md for the full template spec.
devops/
├── .claude-plugin/plugin.json ← Plugin manifest
├── CONVENTIONS.md ← Naming, versioning, extension rules
├── hooks/ ← <!--devops:count:hooks-->35<!--/devops:count:hooks--> hooks (JS) registered in hooks.json
├── skills/ ← <!--devops:count:skills-->22<!--/devops:count:skills--> skill definitions (SKILL.md)
├── agents/ ← <!--devops:count:agents-->12<!--/devops:count:agents--> agent definitions
├── deep-knowledge/ ← Cross-cutting reference docs
├── templates/ ← Output format templates
└── scripts/ ← Utility scripts (build-id, usage)
Claude Code caches plugin marketplace data globally. If claude plugin update reports no update available despite a new version being published, clear the global marketplace cache:
# Windows (Git Bash / WSL)
rm -rf ~/.claude/plugins/cache/dotclaude
rm -rf ~/.claude/plugins/marketplaces/dotclaude
rm -f ~/.claude/plugins/install-counts-cache.json
# macOS / Linux
rm -rf ~/.claude/plugins/cache/dotclaude
rm -rf ~/.claude/plugins/marketplaces/dotclaude
rm -f ~/.claude/plugins/install-counts-cache.jsonThen run claude plugin update devops@Jerry0022 again. Start a new session for changes to take effect.