docs(skills): fix residual self-improvement loop inconsistencies

[rachaelrenk]

Summary

Small follow-up to the docs self-improvement loop work (#239) and the aeo_crosslink_audit reliability fix (#280). A retrospective of the "software factory" surfaced a few residual inconsistencies; this PR cleans them up. No changes to published docs — all edits are confined to .agents/.

Changes

.agents/skills/improve-drafting-skills/SKILL.md

• Stop committing signal logs to protected main. Steps A and B previously instructed committing pr_review_runs.md and human_review_feedback.jsonl directly to main with a silent "if the push fails, continue" fallback — the exact failure mode docs(skills): make aeo_crosslink_audit reliable #280 fixed for aeo_crosslink_audit (protected main rejects the push, so the "durable" log never persists). Both now persist through a standing chore/drafting-signal-logs branch and one perpetual, low-noise PR, mirroring the pattern from docs(skills): make aeo_crosslink_audit reliable #280.
• Added a "Persisting the signal logs" subsection that describes the standing-branch flow once, referenced from Steps A and B.
• Noted in the analysis step that prior-run human feedback lives on the standing branch (read it there, or merge the standing log PR first).
• Removed the leftover self-correcting sentence in the Run log section and its stale style_lint_runs.jsonl reference.

.agents/logs/style_lint_runs.jsonl (deleted)

• Orphaned empty file. Since the design moved to [SIGNAL:style-lint] stdout markers (collected via oz run get), nothing writes this file. Removed to match the current design.

Not changed (intentionally)

• improve-aeo-crosslink-skill cadence. PR docs: implement self-improvement loop architecture #239's description called Loop 4 "quarterly," but the skill file runs monthly (matching the sibling improve-drafting-skills). Confirmed monthly is intended; left as-is.

Validation

• No remaining style_lint_runs references repo-wide.
• No remaining "commit directly to main" instructions in improve-drafting-skills.
• git diff --check clean; skill frontmatter parses.

Conversation: https://staging.warp.dev/conversation/a8bc7071-0fda-470a-8f13-49991355164a

Co-Authored-By: Oz oz-agent@warp.dev

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docs	Ready	Preview, Comment	Jul 2, 2026 7:01pm

[oz-for-oss]

@rachaelrenk

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[oz-for-oss]

Overview

This PR updates the improve-drafting-skills workflow to persist drafting signal logs through a standing branch/PR and removes the obsolete style_lint_runs log. The direction matches the stated goal, but the new standing-branch flow leaves correctness gaps that can still drop signals or mix monthly improvement edits into the log PR.

Concerns

• The persistence helper checks out the standing log branch but never returns to a clean improvement branch before later skill/template edits.
• The analysis step still only names Step A in-memory records plus on-disk human feedback, so Step B's failure fallback can discard current-run human feedback.

Verdict

Found: 0 critical, 2 important, 0 suggestions

Request changes

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[oz-for-oss]

⚠️ [IMPORTANT] The helper checks out chore/drafting-signal-logs but never switches back before the monthly run drafts skill/template edits, so those edits can land on the standing log branch and pollute the log PR.

Suggested change

This produces one perpetual, low-noise PR that accumulates every run's log entries regardless of outcome. Reviewers merge it periodically (at minimum before each monthly run) so the logs reach `main`. If any git step fails, keep the in-memory records for this run's analysis and note the failure in the Slack summary.

This produces one perpetual, low-noise PR that accumulates every run's log entries regardless of outcome. Reviewers merge it periodically (at minimum before each monthly run) so the logs reach `main`. If any git step fails, keep the in-memory records for this run's analysis and note the failure in the Slack summary. After the log update is pushed, switch back to `main` (or create the monthly improvement branch from the latest `origin/main`) before making any skill or template edits so the standing log branch only contains log files.

[oz-for-oss]

⚠️ [IMPORTANT] Step B says to keep accepted human-feedback records in memory if the log-branch git steps fail, but this analysis input only reads human feedback from disk, so the fallback path can drop the current run's human signals.

Suggested change

- **On-disk human feedback** — read `.agents/logs/human_review_feedback.jsonl` line by line (skipping empty lines). Each line is a JSON record; parse and filter to the past 30 days. Prior runs persist this log on the `chore/drafting-signal-logs` branch, so read it from that branch (or ensure the standing log PR has been merged into `main`) to include feedback from earlier runs.

- **Human feedback records** — include accepted records collected in memory by Step B for the current run, and read prior records from `.agents/logs/human_review_feedback.jsonl` line by line (skipping empty lines). Each JSON record should be parsed and filtered to the past 30 days. Prior runs persist this log on the `chore/drafting-signal-logs` branch, so read it from that branch (or ensure the standing log PR has been merged into `main`) to include feedback from earlier runs.