AgentOps

Your AI agent forgets everything between sessions. AgentOps fixes that.

Cross-session memory, validation gates, and orchestrated parallel execution for AI coding agents.

Install

npx skills@latest add boshu2/agentops --all -g

Or: claude plugin add boshu2/agentops

See It Work

Catch risks before you code:

> /pre-mortem "add OAuth integration"

[council] Spawning 2 judges: pragmatist, skeptic
[pragmatist] RISK: Token storage — localStorage is XSS-vulnerable
[skeptic] RISK: Refresh token rotation — silent failure on expiry
Verdict: 3 risks found, 1 critical → fix before implementing

Validate before you commit:

> /vibe

[toolchain] gitleaks ✓  semgrep ✓  shellcheck ✓
[council] Spawning 2 judges...
[complexity] All functions within threshold
Verdict: PASS — no critical findings, ready to ship

Parallelize with fresh context per agent:

> /swarm

[swarm] Creating team with 3 workers...
[worker-1] ✓ implement-auth (2m 14s)
[worker-2] ✓ add-rate-limiter (1m 48s)
[worker-3] ✓ update-tests (3m 02s)
All 3 tasks complete. Run /vibe to validate.

What You Get

• Cross-session memory — learnings persist to .agents/ and inject into future sessions (/inject, automatic with ao CLI)
• Validation gates — multi-aspect code review (security, complexity, architecture, and more) that blocks bad merges (/vibe)
• Parallel execution — fresh-context agents work simultaneously without stepping on each other (/swarm)
• Shift-left risk analysis — simulate failures before writing code (/pre-mortem)
• Progress locks — once a gate passes, it stays passed — no regression (/ratchet)
• Autonomous execution — orchestrate multi-issue work across waves of parallel agents (/crank)

Skills

Skill	What it does
/pre-mortem	Simulate failures before coding
/vibe	Complexity analysis + multi-model validation gate
/swarm	Parallel agents with fresh context
/crank	Autonomous epic execution (orchestrates /swarm waves)
/council	Multi-model consensus — validate, brainstorm, research
/implement	Single issue, full lifecycle
/post-mortem	Extract learnings for future sessions
/research	Deep codebase exploration
/plan	Break a goal into tracked issues
/release	Pre-flight, changelog, version bumps, tag
/ratchet	Progress gates that lock

More skills

Skill	What it does
/retro	Quick retrospective
/inject	Manually load prior knowledge
/knowledge	Query knowledge base
/bug-hunt	Root cause analysis
/complexity	Code complexity metrics
/doc	Documentation generation
/standards	Language-specific coding rules
/beads	Git-native issue tracking

How It Works

Pattern	What it solves
Fresh context per agent	Context bloat degrades performance — each agent starts clean
Validation gates	Work regresses silently — must pass /vibe to commit
Orchestrated execution	Multiple agents cause chaos — one orchestrator owns the loop, agents work atomically
Compounding memory	Same bugs rediscovered — /post-mortem → .agents/ → next session

See docs/reference.md for architecture diagrams, execution modes, and the full pipeline.

Going Deeper

Optional: ao CLI — adds session hooks that auto-inject prior knowledge at session start and auto-extract learnings at session end. Without it, you run /inject manually.

brew tap boshu2/agentops https://github.com/boshu2/homebrew-agentops
brew install agentops
ao hooks install

Optional: beads — git-native issue tracking. Lets /crank orchestrate multi-issue work from a tracked backlog.

See docs/reference.md for per-agent install options, CLI reference, execution modes (local vs distributed), and tool dependencies.

FAQ

Why not just use Claude Code directly? Claude Code has agent spawning built in. AgentOps adds what it lacks:

• Cross-session memory (agents forget everything when the session ends)
• Codified patterns (isolation, validation contracts, debate protocol) that agents won't discover on their own
• Cross-vendor validation (--mixed mode adds Codex judges alongside Claude)

What data leaves my machine? Nothing. All state lives in .agents/ (git-tracked, local). No telemetry, no cloud sync.

How do I uninstall?

npx skills@latest remove boshu2/agentops -g

Built On

These ideas shaped AgentOps — they're baked in, not extra dependencies.

Project	What we use it for
Fresh context per agent	Each spawned agent gets clean context — no bleed-through
Validation gates that lock	Once a stage passes, it stays passed — no regression
beads	Git-native issue tracking
MemRL	Two-phase retrieval for cross-session memory

License

Apache-2.0 · Documentation · Reference · Changelog