kevglynn/house-rules

This skill should be used before implementing features, building components, or making changes. It guides exploring user intent, approaches, and design decisions before planning. Triggers on "let's brainstorm", "help me think through", "what should we build", "explore approaches", ambiguous feature requests, or when the user's request has multiple valid interpretations that need clarification.

Convene a council of diverse AI voices to analyze a task from multiple perspectives using different models. Each voice operates as a full-capability generalPurpose subagent (reads, shell, MCPs) with its own analytical lens. Synthesizes perspectives rather than ranking them. Use when the user says "convene a council", "council on this", "get perspectives", "multi-voice", "diverse review", or wants adversarial/collaborative analysis from multiple models.

Whole-repo audit for over-engineering. Like graybeard-review, but scans the entire codebase instead of a diff: a ranked list of what to delete, simplify, or replace with stdlib/native equivalents. Use when the user says "audit this codebase", "audit for over-engineering", "what can I delete from this repo", "find bloat", "graybeard-audit", or "/graybeard-audit". One-shot report, does not apply fixes.

Harvest every defer: comment in the codebase into a debt ledger, so the deliberate shortcuts and deferrals get tracked instead of rotting into "later means never". Use when the user says "graybeard debt", "/graybeard-debt", "what did we defer", "list the shortcuts", "debt ledger", or "what did we mark to do later". One-shot report, changes nothing.

Quick-reference index for the graybeard family: which skill or rule to use for over-engineering review, repo-wide audit, defer-ledger harvesting, or minimal-implementation mode. One-shot display, not a persistent mode. Trigger: "graybeard help", "use graybeard", "which graybeard skill", "how do I use graybeard", "what graybeard commands".

Code review focused exclusively on over-engineering. Finds what to delete: reinvented standard library, unneeded dependencies, speculative abstractions, dead flexibility. One line per finding: location, what to cut, what replaces it. Use when the user says "review for over-engineering", "what can we delete", "is this over-engineered", "simplify review", or invokes /graybeard-review. Complements correctness-focused review, this one only hunts complexity.

Author a gate/sprint review packet from a committed blueprint: gather the gate's bead evidence and receipts, populate the reader's-own-checklist status grid with a divergence strip, a capped FAQ, and exactly three linked receipts, carry the decision vocabulary into the reader's repo, then verify it headlessly and share the single packet.md markdown file. Use when the user says "build the packet", "gate packet", "sprint packet", "assemble the G2/G3 packet", or when a gate/sprint reaches close and needs a review artifact for a cold reader.

Validate a drafted gate/sprint packet with a four-lane deep-dive: accuracy-vs-repo, cold-read (the overlay's reader persona and access boundary), receipts verification (re-run every count/string/exit-code live), and navigability/structure — with a phase-2 convergence round between accuracy and cold-read, wrapped around the deep-dive-orchestrator subagent, emitting the findings-and-dispositions review doc. Use when the user says "deep-dive the packet", "validate the packet", "packet review", "cold-read the gate packet", or after the packet skill drafts a packet that needs validation before it reaches the reader.

Per-bead-type TDD playbooks — the operational procedure behind the pragmatic-tdd rule. Use when starting a bug/feature/refactor bead, when you need to write the failing test, or when proving red-then-green after the fact. Triggers on "pragmatic tdd", "starting a bug/feature/refactor bead", "write the failing test", "red-proof", "prove the test fails".

Voice discipline for human-facing prose — pattern catalog, self-correction protocol, and worked examples for killing AI-sounding rhetorical tics. Load when writing positioning docs, pre-whitepapers, packets, deck copy, primers, emails, briefs, or any prose a human will read or forward. Trigger phrases: "prose voice", "writing voice", "check the voice", "before I send this", "make it not sound like AI", "voice sweep".

Run a deliberate, evidence-grounded audit of a project's intent, plan, execution, and operating mechanisms, producing a classified refinement proposal — a structured recommendation for change, or an explicit confirmation that no change is needed. Covers active-plan and retrospective modes, any scope from full project down to a single policy, and mechanism-placement audits ("is this capability on the right surface?"). Use when the user says "run a refinement", "refine this epic", "refine this plan", "refine this rule", "is this still the right shape", "mechanism placement audit", or "retro on this capability".

Structured approach to diagnosing and fixing bugs. Use when a bug is reported, a test fails unexpectedly, behavior doesn't match expectations, or when "something is wrong" but the cause isn't clear. Triggers on "debug this", "why is this broken", "this doesn't work", "investigate this failure", unexpected test failures, or when multiple fix attempts have failed.

Run a Tier 1 same-model multi-lens review of a completed chunk of work: dispatch independent review subagents (code-reviewer, architect-reviewer, simplify, and — when the change involves tests or a no-test claim — test-signal) in parallel with a complete five-element context package, triage findings into Critical/Important/Minor with a disposition per finding, land fixes as a separate commit, file beads for deferred minors, and emit a close-evidence line. Use when the user says "tier 1 review", "run tier 1", "three-lens review", "multi-lens review", or when a feature/refactor bead reaches the self-review-done stage and the change meets the multi-agent-review trigger criteria.

Run a Tier 2 cross-model review handoff on a Tier-1-reviewed chunk of work: gather the bead context, changed files, Tier 1 outcomes, and trusted boundary; draft the review-focus paragraph; call the deterministic assembler to emit a ready-to-paste prompt document for external models; then triage the returned findings (accept/reject with reason, conservative on disagreement) and emit the close-evidence line. Use when the user says "tier 2 review", "cross-model review", "tier 2 handoff", "assemble the review prompt", or when a shared-primitive bead has passed Tier 1 and needs external-model eyes before close.

Orients an agent in a new or resumed workspace by reading the canonical orientation files (CLAUDE.md, AGENTS.md, .cursor/scratchpad.md, and PLAN.md if present), summarizing what the workspace is for, current status, and the top blockers and next actions — then waiting for direction instead of executing. Use when the user says "start this workspace", "resume this workspace", "kick it off", "begin session", "pick up where we left off", "what is this workspace", or opens a scaffolded workspace for the first time in a session.