Agent
Curates the host's durable context — root CLAUDE.md, collection-level CLAUDE.md files under / /, and the current engineer's per-user memory tree under /.claude/projects/ /memory/. Use when capturing a non-obvious pattern, a post-incident learning, a behavior correction, or promoting a personal memory into team…
Agent
General correctness/security/platform adversary for milestone-pipeline v2. Walks the 11 workspace institutional-memory axes on the implementation diff and emits a severity-tagged critique. Always fires independently alongside the always-on delivery-integrity adversary; frontend and infra critics remain conditional.…
Agent
Independent read-only closure reviewer for milestone-pipeline v2. Runs after rectification and before implementation-evidence assembly; verifies every C/H disposition, review hashes, raw check evidence, and the final head. It cannot implement or rectify.
Agent
Always-on blind adversary for milestone-pipeline v2. Tries to falsify implementation, review, publication, plan, authorization, apply, verification, waiver, freshness, replay, and append-only claims. Reviews independently from the general milestone adversary and emits V-prefixed findings. Read-only; never rectifies…
Agent
Conditional frontend/UI/UX critic for milestone-pipeline v2. Fires for frontend paths and reviews visual hierarchy, states, copy, reflow, theme parity, and accessibility with F-prefixed findings. It supplements both always-on adversaries and works blind from sibling critiques. Read-only except for its assigned…
Agent
Implementation agent for the milestone-pipeline. Reads the merged research briefs and produces commits on the appropriate source branches. Used for the delegated-implementation path (larger-than-inline scope or explicit parallel-explorer worktrees). For specialist-shaped work (helm-apps, gitops, service-mesh…
Agent
Conditional infrastructure-safety critic for milestone-pipeline v2. Fires for high-risk infra repo identity or paths and reviews Pulumi, IRSA, PCA, Istio, ArgoCD, network blast radius, authorization, observability, and rollback with I-prefixed findings. It supplements both always-on adversaries and works blind from…
Agent
Independently audits a published release and frozen operations plan before any live apply authorization.
Agent
Exception-path rectification agent for milestone-pipeline v2. Re-verifies every finding, fixes C/H and small M, records dispositions, reruns checks, and commits locally. It never publishes or self-certifies code-complete; an independent closure verifier runs afterward.
Agent
Research agent for the milestone-pipeline. Produces a research brief covering in-codebase context (workspace MCP, charts/, infra/, source/) and external context (vendor docs, OSS landscape, papers). Dispatched in parallel (typically 2× per pipeline run) to surface diverse approaches. Returns only the brief path +…
Agent
Phase 2 DECOMPOSE agent for the /roadmap pipeline. Converts the goal statement into 2–6 epics (≤ 6 weeks each) with INVEST validation, enabler/value tagging, specialist-agent citation, and epic-level acceptance criteria. Writes the ## Epics section to the roadmap doc. Inputs: {SLUG}, {ROADMAPPATH}. Dispatched by the…
Agent
Phase 4 MATERIALIZE agent for the /roadmap pipeline. Finalizes the roadmap doc by writing the ## Cross-references section, linting via roadmap-validate.py, emitting + validating the machine-readable milestone register (plans/ -milestones.json), and (if GITLABMODE=draft-only) drafting GitLab issue bodies to local files…
Agent
Phase 1 REFINE agent for the /roadmap pipeline. Converts a vague brief into a 1-page goal statement with HMW reframe, sharpening questions, assumption tiers, Objective + KRs, and Won't list. Writes the ## Goal section to the roadmap doc. Inputs: {SLUG}, {BRIEF}, {ROADMAPPATH}. Dispatched by the /roadmap slash command…
Agent
Phase 3 SEQUENCE agent for the /roadmap pipeline. Orders epics into Now/Next/Later lanes via MoSCoW + RICE (calling roadmap-score-moscow.py and roadmap-score-rice.py), decomposes Now-lane epics into stories ≤ 3 days each, assigns milestone IDs, and builds the spike/discovery lane. Writes the ## Roadmap — Now / Next /…
Agent
Turn 2 author for the /skill-to-command pipeline. Writes the new slash command body + phase subagents + migrated references/scripts on a designated write surface during the parallel-authoring phase. Dispatched TWICE in parallel by the /skill-to-command orchestrator — Agent A covers orchestrator+agents+references…
Agent
Turn 3 adversarial reviewer for the /skill-to-command pipeline. Walks the 14 critical gotchas catalog against the newly authored tree (data/commands/ .md + data/agents/ -.md + migrated refs/scripts + deprecated stub) and emits a severity-tagged critique with CRITICAL/HIGH/MEDIUM/LOW findings. Dispatched ONCE by the…
Agent
Phase 1 designer for the spike pipeline. Extracts the spike brief from the roadmap and designs the experiment as a TYPED contract — writes design.json (assumption, criteria with field/operator/threshold/unit, sample-size justification, ≥3 confounds, measurementfields, poc constraints, cost) plus the human design.md…
Agent
Phase 2 executor for the spike pipeline. Reads design.json (the typed contract) + design.md, implements the POC (≤200 LOC stdlib-only Python), runs measurements, and writes measurements.json — one value per measurementfield, echoing the orchestrator-provided designhash, declaring any unmeasurable field explicitly …
Agent
Phase 4 adversarial reviewer for the spike pipeline. Reads all artifacts independently, forms its own verdict from the raw data BEFORE reading note.md, then applies 6 validation axes and writes review.json (typed — six axes + an ACCEPT/RE-RUN/RECONSIDER-DECISION verdict, echoing the checkpointed decisionhash+notehash)…
Agent
Phase 3 writer for the spike pipeline. Reads decision.json (the DERIVED verdict), design.json, measurements.json, and poc/, and writes note.md following the canonical template. The writer does NOT choose the verdict — spike-decide.py already derived it; the writer cites it verbatim and writes decisive, imperative…
Command
Promote a durable agent-memory lesson to the right tier — repo wiki Field Notes (gated), repo CLAUDE.md, or data/references — without duplicating it across tiers. Use when a pipeline agent captured a significant, durable lesson in .claude/agent-memory/ /lessons.md that is useful beyond that one agent/run, or when the…
Command
Run the canonical delivery-state v2 milestone pipeline (Research → Implement → independent adversarial review → Rectify/closure → Publish → Apply → Verify). Use for a non-trivial fixed milestone id when code/checklist completion must be distinguished from live operational delivery.
Command
Build a structured technical roadmap (Goal → Epics → Now/Next/Later milestones → Stories) from a problem statement or the current conversation. Use when the user invokes /roadmap , says "plan this project", "build a roadmap for", "decompose this into milestones", "what's the path to ship X", or has been discussing a…
Command
Convert an existing skill at data/skills/ /SKILL.md into a slash-command + N-subagent + agent-memory pattern. Use when a skill has grown into a multi-phase orchestrator (e.g. has "Phase 1 / Phase 2 / Phase 3 / Phase 4" structure, dispatches sub-tasks, holds state across phases) and you want to refactor it to match the…