Scrub Reflection Self Improvement
Developmentscrub-reflection-self-improvementScrub Reflection Self Improvement is an Agent Skill for development published in the marin-community/marin repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Scheduled scrub: repository self-improvement.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/marin-community/marin/blob/HEAD/.agents/skills/scrub-reflection-self-improvement/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/scrub-reflection-self-improvement, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
scrub-reflection-self-improvement
Use this skill on scheduled scrub turns to identify and land high-leverage improvements in marin-community/marin.
Focus
- Look for improvements from recent issues, PR feedback, and recurring operational friction.
- Prefer one concrete implementation per run when feasible.
- If implementation is blocked, produce a concrete plan and capture follow-up work in GitHub.
Candidate Signals
- Repeated confusion in docs, recipes, or contributor workflows.
- Recurring failures or avoidable manual steps in experiments, scripts, and infra operations.
- Capability gaps that reduce the value of agent-assisted contributions.
Triage Checklist
Run a lightweight, repeatable scan before choosing work:
- Review recent open issues and open PRs for repeated friction clusters.
- Explicitly check for already-open scrub-generated issues/PRs touching the same area; prefer advancing or deferring to that existing artifact instead of creating a parallel one.
- Review the latest commits on
mainfor changes that imply follow-on docs/workflow updates. - Search
AGENTS.md,.agents/skills/, and docs for stale workflow guidance related to those clusters. - De-duplicate against existing issues/PRs before creating new artifacts.
When possible, prefer improvements that remove recurring operator time (for example, turning ad-hoc scrub judgment into explicit repeatable guidance).
Decision Heuristics
- Pick the highest-leverage change with the lowest coordination overhead.
- Treat open scrub-generated issues/PRs as first-class prior art during triage; if one already covers the candidate improvement, avoid opening a second artifact unless the new scope is clearly distinct.
- De-duplicate against existing issues/PRs before opening new work.
- When an improvement changes recurring workflow guidance, codify it in durable repo instructions:
AGENTS.mdfor cross-cutting agent behavior, or.agents/skills/for repeatable task workflows. - If no justified improvement exists now, choose a no-op outcome.
- Prefer direct implementation over opening new issues when the change is fully in-repo and low-risk.
Output
- Keep rationale explicit: observed gap, change made (or plan), and expected impact.
- Prefer durable artifacts over transient notes: land guidance updates in
AGENTS.mdand/or recipe docs when that is the primary improvement. - Treat local-only edits as incomplete work. If you modify files, publish the result (commit/push and open or update a PR per
.agents/skills/commit/SKILL.md) before finishing this scrub run. - If publish is blocked (auth, permissions, CI infra, etc.), report the blocker and set a future
needs_followup_atinstead of ending the run. - If you choose no-op, include explicit inspected signals and why no justified improvement exists now.
- End the run with exactly one footer line of valid one-line JSON:
HARNESS_SCRUB_LOOP {"needs_followup_at":null}. Setneeds_followup_atto null when the run is complete, or a future RFC 3339 timestamp when another follow-up turn is needed.
For no-op outcomes, include at minimum:
- which issue/PR/commit windows were inspected,
- why each candidate was not suitable for this run,
- and why deferring to the next scheduled run is preferable to opening a low-signal artifact now.