CtrlK
BlogDocsLog inGet started
Tessl Logo

sweep

Post-op check for artifacts, damage, and stale references after agent work

60

Quality

70%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./skills/sweep/SKILL.md
SKILL.md
Quality
Evals
Security

Post-Op Sweep

Overview

Audit the working tree after agent work completes. Find what the agents left behind.

1. Blast Radius

Map every file touched, across all repos affected:

git diff --stat
git diff --cached --stat

If other repos were modified (check recent tool output or plan files for cross-repo work), run git diff --stat there too.

Flag anything unexpected, files not mentioned in the plan or task.

2. Fixture and Example Damage

Check whether test fixtures, example files, seed data, or sample configs were modified:

  • Look for changes to files with names like example, fixture, seed, sample, mock, test-data
  • Changes that null out values, remove content, or alter structure are likely test side-effects
  • These should almost always be reverted

3. Stale References

Search for references that became outdated by the work:

  • Plan files referencing work as "pending" or "TODO" that's now done
  • CLAUDE.md / README listing features as planned that are now implemented
  • Comments saying "not yet implemented" for code that now exists
  • Config or docs pointing to old file paths, function names, or APIs that changed

4. Debug Artifacts

Scan for artifacts agents leave behind:

  • console.log, print(), echo "DEBUG" statements
  • Temporary files (*.tmp, *.bak, /tmp/ references)
  • Commented-out code blocks that serve no purpose
  • Test data written to production data directories

5. Scope Bleed

Identify changes outside the task scope:

  • Unrelated formatting changes
  • Dependency updates not requested
  • Refactors beyond what the plan called for
  • Changes in other repos' working trees that predate this session

6. Plan Criteria

If plan files exist for the work just completed, check acceptance criteria:

  • Read each criterion
  • Verify whether it was met (by reading code, not re-running tests)
  • Flag any unmet criteria

Report

Present findings as a short checklist:

  • Clean, no issues found (say so and stop)
  • Fix, items that need correction (damaged fixtures, stale refs, debug artifacts)
  • Note, items worth awareness but not blocking (scope bleed, unrelated changes)

Be terse. Name files and line numbers. Skip categories with no findings.

See Also

The post-agent triad: sweep checks for collateral damage, the other two check fidelity and threads.

  • /debrief: cross-session recap of what agents accomplished and the loose ends
  • /agent-watchdog: audit whether an agent did what was asked, with a gap report and a narrow fix
Repository
tslateman/duet
Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.