Diagnose hard bugs and performance regressions with reproducible evidence, focused hypotheses, and proportional instrumentation. Use when something is broken, failing, throwing, intermittently wrong, or unexpectedly slow.
The English summary is being prepared. Showing the original description.
See how to installReview the instructions the agent will receive before you install.
Start from evidence and adapt the depth of the investigation to the problem. Read relevant repository context and architectural decisions when they exist.
Reproduce the reported symptom with the cheapest signal that distinguishes broken from fixed. A focused test or script is ideal when practical; logs, traces, snapshots, comparisons, or measured timings may be better for other failures. Tighten the loop by improving speed, specificity, and determinism.
If an exact reproduction is unavailable, continue with the strongest evidence available and state the limitation. Request an artifact, access, or temporary instrumentation only when it would materially improve the diagnosis.
For feedback-loop options and debugging tactics, load the
debugging playbook as needed. For a rare
manual reproduction, adapt
scripts/hitl-loop.template.sh.
Minimize the reproducing scenario when doing so will shrink the search space. Form a small ranked set of falsifiable hypotheses from the evidence, then choose probes that best distinguish them. Share hypotheses with the user when their domain knowledge could redirect the investigation; do not make routine progress depend on a checkpoint.
Use targeted instrumentation at boundaries that separate plausible causes. Change as little as practical per probe. For performance regressions, establish a baseline and use profiling, query plans, or bisection before optimizing.
Add a regression test when a stable seam can represent the real failure. If it cannot, explain the coverage gap rather than adding a misleading test. Apply the smallest justified fix, then rerun both the focused signal and relevant nearby checks.
Remove temporary instrumentation and artifacts unless the user wants to retain them. Report the observed cause, the evidence that supports it, what was verified, and any remaining uncertainty. Recommend architectural follow-up only when the diagnosis exposes a concrete recurring weakness.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Create or review an AGENTS.md file so coding agents get stable repo-local instructions: environment setup, testing, style, security boundaries, PR policy, and handoff rules. Use when a repo lacks durable agent guidance or when a custom harness needs a predictable context file.
The English summary is being prepared. Showing the original description.
Recommend the smallest Matt Pocock skill or short skill sequence that fits the user's current goal.
The English summary is being prepared. Showing the original description.
Decision guide for delegating to caveman-style subagents. Tells the main thread WHEN to spawn `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit), or `cavecrew-reviewer` (diff review) instead of doing the work inline or using vanilla `Explore`. Subagent output is caveman-compressed so the tool-result injected back into main context is ~60% smaller — main context lasts longer across long sessions. Trigger: "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", "compressed agent output".
The English summary is being prepared. Showing the original description.
Ultra-compressed communication mode. 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 /caveman. Also auto-triggers when token efficiency is requested.
The English summary is being prepared. Showing the original description.
Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman.
The English summary is being prepared. Showing the original description.
Ultra-compressed commit message generator. 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", "/commit", or invokes /caveman-commit. Auto-triggers when staging changes.
The English summary is being prepared. Showing the original description.