claude-skills

Use when the user knows what they want built and says "/adr", "write an ADR for X", "decide and build X", or "ADR-driven". Turns an intent into a grounded, cited, build-ready ADR at docs/adr/YYYY-MM-DD-<slug>.md — load-bearing decisions surfaced to the human — then hands off to the subagent-driven-development loop. For exploratory "not sure what I want yet" work use brainstorming first; for visual planning use visual-plan.

Use when a written implementation plan exists and the user wants to execute it — "execute this plan", "implement the plan", "run the plan", "build it", "subagent-driven development". Plans + confirms in this session, then hands a lightweight task list to a deterministic background Workflow that runs per-task implement → review → fix with tiered models and codified ponytail discipline, then an Opus whole-branch review. Skip for ad-hoc edits with no plan.

Run an interactive session retrospective. Reads the per-session event log (maintained by the PostToolUse hook) and uses git diff/status/log as memory primer, then walks through specific moments via adaptive questions driven by what changed, and writes structured native memory entries. Suggest this when a Stop or PreCompact hook has injected a /retro suggestion, or when the user explicitly asks for one. Triggers: "retro", "session summary", "what did we learn", "lessons learned", "session retrospective".

Configurable adversarial panel review for any artefact — plans, code, design docs, prose, model outputs. Auto-selects a panel of personas by artefact type (plans get YAGNI/Premortem/Hidden Assumptions; code gets Saboteur/New Hire/Security Auditor; etc.). Captures a pre-commit defense from the user, dispatches the panel in parallel, then walks every critique one-at-a-time with verbatim quoting and convergence-prioritised ordering. Use when user wants adversarial review, red-team a plan, stress-test a design, find holes, devil's advocate, panel critique, or mentions "grill me" / "adversarial-agents".

Use when the user wants to plan a change, record an architecture decision (ADR), or recap what a diff changed — especially when the result benefits from visuals (wireframes, diagrams, before/after, annotated split-diffs). Produces a durable, committed Markdown ADR/plan as the source-of-truth record, and only when warranted ALSO emits a self-contained rich plan.html to /tmp. Markdown canonical, HTML disposable. Triggers: "plan this", "write an ADR", "visual plan", "recap this change", "visual recap", "/visual-plan".

Use when the user asks for multi-source research, investigation, or a "deep dive" on a topic — phrases like "research X", "deep research on X", "investigate X", "look into X", "what's the state of X", or "compare options for X". Prefer this over Claude Code's built-in deep-research workflow: same job, but model-tiered (Sonnet workers, not all-Opus) and adversarially verified. Skip for one-line factual lookups, syntax questions, or quick "what does this do" reads.