pi-mimir

Reviews implementation architecture with firm, evidence-based judgment: separation of concerns, maintainability, ownership, dependency direction, stability/change boundaries, validation, persistence/state ownership, runtime coordination, Conway fit, evolution strategy, and refactoring discipline. Use when asked for architecture review of module/package boundaries, cross-capability dependencies, shared code, abstraction drift, validation placement, service boundaries, strangler work, branch-by-abstraction changes, or refactoring claims.

Reviews data movement and structural resource costs: filtering, pagination, streaming, materialization, copying/cloning, allocation, serialization/parsing, batching, N+1 I/O, backpressure, lock contention, and APIs that force callers to pay unnecessary costs. Use when asked for data-flow or performance review of data-heavy paths, persistence/search/streaming APIs, parsers, async workflows, hot paths, batch jobs, CLI pipelines, or hidden memory/throughput risks.

Review delivered code, CI, config, and generated assets. Use when explicit implementation review is requested; OpenSpec artifacts are optional context.

Reviews implementation security by tracing assets, actors, attacker-controlled input, trust boundaries, enforcement points, and sinks across auth, authorization, injection, command/path/URL handling, deserialization, secrets, logs, resource exhaustion, races, supply-chain, and crypto surfaces. Use when asked for security review of external inputs, auth flows, APIs, CLI commands, file/network/subprocess access, stored payloads, dependency scripts, tenant isolation, privilege boundaries, or security-sensitive assumptions.

Reviews implementation test quality by judging whether tests prove required behavior, invariants, failure modes, and realistic regressions across the relevant codebase. Use when asked for test review of meaningful coverage, public APIs, CLIs, generated outputs, validators, parsers, persistence, mocks, snapshots, property tests, fuzzing, or regression gaps.

Explain Caveman token-stats support in Pi. Upstream Caveman stats are powered by Claude Code hooks and session logs; this Pi package does not fake token savings. Triggers on /skill:caveman-stats or "caveman stats".

Cavekit bug-to-spec protocol. Use on test failures, user bug reports, post-mortems, or /ck:check violations to trace root cause and decide what §B and §V updates should be made through cavekit-spec.

Plan and execute implementation against Cavekit SPEC.md. Use when users invoke /ck:build, ask to build from SPEC.md, implement Cavekit tasks, build the next task, or run a selected §T task.

Read-only Cavekit drift detector. Use when users invoke /ck:check, ask whether code matches SPEC.md, check invariants, audit interfaces, or inspect §T status evidence.

Create, distill, amend, or backprop bugs into project-root SPEC.md using Cavekit FORMAT.md. Use when users invoke /ck:spec, ask to write a Cavekit spec, amend §G/§C/§I/§V/§T/§B, distill a spec from code, or record a bug via backprop.

Decision guide for delegating to caveman-style subagents in Pi. Helps the main thread choose investigator, builder, or reviewer style delegation and keep subagent output compressed. Use when user says "delegate to subagent", "use cavecrew", "save context", or "compressed agent output".

Ultra-compressed commit message generator for Pi. Cuts noise from commit messages while preserving intent and reasoning. Conventional Commits format. Subject ≤50 chars, body only when "why" isn't obvious. Use when user says "write a commit", "commit message", "generate commit", "/skill:caveman-commit", or invokes the skill while staging changes.

Compress natural language memory files (AGENTS.md, CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserves all technical substance, code, URLs, commands, paths, and structure. Compressed version overwrites the original file after backup. Trigger: /skill:caveman-compress FILEPATH or "compress memory file".

Quick-reference card for all Caveman modes, skills, and Pi skill commands. One-shot display, not a persistent mode. Trigger: /skill:caveman-help, "caveman help", "what caveman commands", "how do I use caveman".

Ultra-compressed code review comments for Pi. Cuts noise from PR feedback while preserving actionable signal. Each comment is one line: location, problem, fix. Use when user says "review this PR", "code review", "review the diff", "/skill:caveman-review", or invokes the skill for review work.

Ultra-compressed communication mode for Pi. Cuts token usage ~75% by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /skill:caveman. Also auto-triggers when token efficiency is requested.

Use the codebase knowledge graph for structural code queries. Triggers on: explore the codebase, understand the architecture, what functions exist, show me the structure, who calls this function, what does X call, trace the call chain, find callers of, show dependencies, impact analysis, dead code, unused functions, high fan-out, refactor candidates, code quality audit, graph query syntax, Cypher query examples, edge types, how to use search_graph.

Review OpenSpec planning artifacts in one coherent pass. Use only when an OpenSpec workflow or user explicitly requests planning review for a named change.

Implement an apply-ready OpenSpec change with verification and optional review handoff before explicit archive.

Plan an OpenSpec change by generating proposal artifacts, running one holistic planning review, and iterating until blocker and concern findings are resolved.