starfleet-claude-code

Scan for stale memory topics, near-dead instincts, and low-salience K-LEAN entries. Proposes cleanup actions interactively.

Quality gate for instinct persistence. Validates new instincts through a 4-stage pipeline before allowing them to be written to disk. Invoke before writing any new instinct file.

Use when the user needs thorough research on a complex topic, says 'deep dive', 'research this thoroughly', or 'sensor sweep'. Parallel agent orchestrator for comprehensive multi-angle investigation.

Autonomous experiment loop: iteratively improves a single file against a measurable metric. Uses git-as-state-management in a disposable worktree. Invoke when user wants to optimize, tune, or systematically improve a file through repeated iterations.

Use when you need multi-faction review combining security, strategy, cost, and quality perspectives. Trigger: fleet review, multi-faction assessment, comprehensive quality gate, or when escalation requires L5+.

Two-loop research orchestrator: chains deep-research (outer, hypothesis generation) with experiment-loop (inner, empirical testing). Use when research questions have a testable dimension with a measurable metric.

Rigorous, citation-backed scientific research with evidence grading and claim-evidence tracing. Use when research needs academic-grade citations, source verification, or systematic review methodology. Trigger: /scientific-research, 'research with citations', 'find papers on', 'systematic review', 'literature review'.

Use when (1) you want an independent review from a different model, (2) stuck or looping on an issue, (3) need cross-model adversarial critique, (4) user says 'ask codex', 'ask gemini', or 'get a second opinion'. Cross-model collaboration with OpenAI Codex and Google Gemini CLI.

Use for significant code changes requiring multi-reviewer adversarial cross-challenge. Trigger: adversarial debate, cross-review, or when quality gates require L6 escalation.

Structured browser QA with health scoring (0-100), fix cycles, and regression tests. Use when testing a deployed site, running QA, or user says 'away mission qa' or 'test the site'.

Use for complex implementation tasks requiring coordinated multi-agent execution. Trigger: away team, team up, multi-agent implementation, or tasks too large for a single agent.

Metacognitive know/don't-know assessment before deep research or complex implementation. Surfaces knowledge gaps to prevent wasted exploration cycles.

Restrict edits to specific files/directories. Advisory warnings on out-of-scope changes. Use when focusing on a specific area, or user says 'containment field', 'freeze', 'guard', or 'restrict scope'.

Promote patterns into instincts, detect skill graduation candidates, and export/import instincts across projects. Trigger: evolve, promote patterns, update instincts, export instincts, import instincts.

Post-deploy canary: baseline capture, regression detection, relative comparison. Use when verifying a deploy, checking for regressions, or user says 'long range sensors', 'canary', or 'check deploy'.

Release pipeline: test, review, version, changelog, commit, PR, CI verify. Use when shipping a feature, releasing code, or user says 'make it so' or 'ship it'.

Multi-phase security health check for applications. Runs static analysis, infrastructure hardening, and live probing with fleet-powered adversarial review. Use for security assessments of own apps or client projects.

Context exhaustion recovery protocol. When an agent hits hull integrity Red, creates a turnover brief and spawns a replacement to continue the work.

Structured session retrospective measuring skill accuracy, governance friction, and instinct proposals. Invoke after L3+ quality gate work or significant implementation sessions.

Hook health check and test runner — verify hooks work, hook performance, diagnose hook failures. Trigger: hook health check, verify hooks, test hooks, hook performance.

5-officer multi-perspective deliberation for architecture, risk, and major feature decisions. Use for significant design choices that benefit from diverse viewpoints.

Use for cost optimization, token usage analysis, or ROI assessment of skills and workflows. Trigger: cost audit, token savings, caching opportunities, ROI analysis, or 'is this worth the tokens'.

Use for security-focused red team review of code, APIs, or infrastructure. Trigger: security review, red team, attack surface, vulnerability analysis, or when quality gates require L4 for auth/crypto/secrets.

Invoke the Devil's Advocate for steel-man + 10-dimension attack on any proposal, code design, or plan. Confidence-scored findings with constructive suggestions.

Use for strategic analysis of projects, competitive positioning, or hidden risk detection. Trigger: strategic review, competitive analysis, opportunity assessment, or 'what are we missing'.

Invoke historical/fictional expert personas for deep domain analysis. Use when you need expertise-based reasoning beyond standard faction review -- first-principles questioning, deductive debugging, strategic analysis, experimental design, or complexity reduction.

Use when (1) task involves investigation + implementation, (2) 3+ files need changing, (3) user says 'check your approach', or (4) task contains research-then-fix patterns. Prevents wrong-approach errors via adversarial self-challenge before implementation.

Unified self-learning cycle that assimilates patterns from learning queues, reflect, counselors-log, and K-LEAN into the knowledge graph. Cross-references all learning sources for consensus-based pattern extraction.

Analyze observation and learning queues, cluster patterns, propose instincts with confidence scores, and suggest capability evolution when patterns cluster.

Adjust reasoning depth and workflow behavior — quick/standard/thorough profiles. Trigger: go faster, be more thorough, reasoning depth, effort level.

Run 4 lightweight quality checks on recent work and fix issues found

Reviews captured learnings from the learnings queue and applies them with user approval. Invoke with /reflect to process queued corrections, preferences, and positive feedback from recent sessions. Use after several sessions to consolidate learning into CLAUDE.md and memory files.