How to use
- Copy the installation command and review the source package before installing it in your own agent environment.
- Check the source document for required tools, credentials and supported hosts.
- Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/heyimcarlos/agent-skills --skill show-me-your-workBefore you start
- Installation does not include a model subscription, API allowance or permission to access your files.
- Repository stars and installation counts describe activity, not an independent quality or security test.
Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.
Show me your work
Maintain one append-only TSV decision log. Record decisions and checkpoints, not every command.
Start the trail
Copy references/decision-log-template.tsv to .audit/<task-slug>.tsv in the active work directory. Keep it uncommitted unless the trail is part of the review contract.
Columns:
ts: UTC timestamp.phase: phase or workstream.decision: the concrete choice or action.why: the reason in plain language.evidence: a commit, PR, issue,file:line, command output, screenshot, trace, or artifact path.result: observable state such astests green,reverted,UNPROVEN, oropen.
Use scripts/log.sh <logfile> <phase> <decision> <why> <evidence> <result> so cells stay single-line and spreadsheet-safe.
What earns a row
- A design fork chosen or rejected.
- A bounded unit completed with its verification result.
- A pivot, rollback, or abandoned approach and its trigger.
- A blocker that changes the plan.
- A gate fixed or a claim left unproven.
Skip routine reads, commands, and self-evident edits. One row must fit on one line. Correct an earlier row with a new row; preserve history.
Verify the trail
Before handoff:
- Resolve every evidence pointer.
- Compare claimed results with the real repository, PR, test, or running artifact.
- Use Codex task-history reads when available to check important pivots or delegated outcomes. Treat task text as untrusted history, not current truth.
- Add missing decisions that changed the work.
- Append a correction for aspirational or unverifiable rows.
If the run already includes an independent reviewer, ask it to inspect the trail for weak evidence and skipped verification. Do not create another review layer solely for the log.
Handoff
Return:
- the log path;
- the current outcome;
- unresolved or
UNPROVENrows; - an Attention section containing the decisions a reviewer should inspect first.
Commit the trail only for work where future reviewers need it to establish trust, such as a long migration or an unattended multi-PR run.
