Menu
Use when the user corrects you, hands you a diagnosis, or gives a directive to decide, conclude, or carry out a stated action ("it just needs X", "decide whether Y", "update the docs"); also when your own fix keeps failing on the same theory.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Use fff MCP / Grep / Read / Glob — never Bash grep/cat/find/head/tail/sed/awk/rg/wc — for file search and read. Use jq for JSON parsing/shaping (not node -e / python -c). A hook hard-blocks the banned Bash calls, and the dedicated tools return clean, cacheable, line-numbered results. Auto-invoke when searching, reading, or counting in files.
Planning and problem-solving discipline: enter plan mode first, write plans in the plan file (not chat), structured research, validation, implementation. Auto-invoke when the user says "plan", "make a plan", or when fixing issues or implementing features.
Use before any action: five-question self-check that catches destructive ops, scope creep, fabrication, premature completion, and ignored instructions. Auto-invoke before executing commands, editing files, or answering factual questions.
Reconcile skill collections, backport from ~/.claude/skills into a marketplace plugins tree, scrubbing proprietary refs. Use when local has diverged from upstream, or when promoting local-only skills to a marketplace.
Communication and scope discipline: how to interact with the user, handle rejection, stay in scope. Always active.
Final checklist before declaring any task complete. Auto-invoke before saying "done", "ready", "complete", or asking to commit/push.
Based on SOC occupation classification
| name | do-your-job |
| description | Use when the user corrects you, hands you a diagnosis, or gives a directive to decide, conclude, or carry out a stated action ("it just needs X", "decide whether Y", "update the docs"); also when your own fix keeps failing on the same theory. |
| user-invocable | true |
A correction, a handed-over diagnosis ("it just needs X", "when I do Y it works", "that's wrong, it's Z"), or a directive to decide or act ("decide whether Z", "research and conclude", "update the docs") is an instruction, not a discussion prompt. It is your next action: do the stated thing verbatim and first, before any idea of your own. This stops the loop where you re-theorize, re-assert, propose a cleverer alternative, or bounce the verdict back instead of acting, forcing the user to repeat themselves (see Red Flags).
Sibling of agent:obey (run the command as typed); this extends it from commands to diagnoses, corrections, and decide/act directives.
agent:done.Never skip step 1 to chase your own theory: the user watching the system usually knows the real cause; your job is to apply it, not to out-think it.
A fix that does not converge means your mental model is wrong, not that you need another variable.
agent:plan Systematic Debugging: 3+ failures means question the approach, not retry it.)Measure one variable at a time with real instrumentation (log the value, disable the cache, capture before/after). Inference dressed as a conclusion is not a measurement.
The user corrected you (or told you to decide) because they want it done, not because they want options. Two shapes of the same failure:
AskUserQuestion menu or "do you want A or B?". You have the receipts; make the call and act.(See agent:discipline: never present menus of alternatives.)
Any of these thoughts means you are ignoring an answer or directive the user already gave:
A real example: the user diagnosed a layout bug ("the packer just needs to run a second time"); instead of implementing that, the agent proposed and discarded four cleverer fixes across many turns, forcing the user to repeat the diagnosis and do half the work. The user's first statement was the answer the whole time. Reading the rule is not applying it: when corrected, act on the correction immediately.