docs(skills): improve drafting skills from signal log patterns 2026-06-29#266
docs(skills): improve drafting skills from signal log patterns 2026-06-29#266rachaelrenk wants to merge 8 commits into
Conversation
…2026-06-29 Co-Authored-By: Oz <oz-agent@warp.dev>
…6-29 Add two additive formatting rules to draft_docs step 6.5, both backed by recurring human review feedback on agent-authored PRs: - Descriptive, specific headings (PRs #143, #189) - Tables/parallel bullets for comparison and reference data (PRs #143, #160, #166) Co-Authored-By: Oz <oz-agent@warp.dev>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
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 Powered by Oz |
There was a problem hiding this comment.
Overview
This PR adds the June 29 drafting-skill improvement signal log and updates draft_docs critical formatting guidance with two additive rules for specific headings and comparison/reference-data structure.
Concerns
- No blocking correctness, security, or spec-alignment concerns found in the annotated diff.
spec_context.mdreports that no approved or repository spec context was available for this PR.
Verdict
Found: 0 critical, 0 important, 0 suggestions
Approve
Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).
Powered by Oz
oz run get (no flags) returns only RunItem.status_message — a brief status text — not full conversation content or shell stdout. Markers printed via echo were invisible to the improve-drafting-skills collector. Fix: emit [SIGNAL:*] markers as agent text output (TextContentBlock) rather than via shell echo, and update the collector to use oz run get --conversation to scan assistant messages. Also fixes the same pattern in review-docs-pr. Co-Authored-By: Oz <oz-agent@warp.dev>
…afting-skills
oz run get --conversation returns {steps: [...]}, not {messages: [...]}.
Steps can be nested, so use recursive descent to reach all assistant
messages at any depth. Added a concrete jq example.
Co-Authored-By: Oz <oz-agent@warp.dev>
…ption Adds check_pr_body.py, a reusable checker that detects repetition-loop degeneration (repeated spans, truncated inline code) and missing/duplicate headings in a PR body before it is submitted. Wires it into create_pr (create + safe additive-edit flows) and improve-drafting-skills Step 7. Co-Authored-By: Oz <oz-agent@warp.dev>
Monthly
improve-drafting-skillsouter-loop run for 2026-06-29. Reads the last 30 days of signal data, identifies recurring drafting errors, and applies targeted, additive edits to the drafting skills.Patterns addressed
Both edits land in
draft_docs/SKILL.mdstep 6.5 (Critical formatting rules), the cross-cutting "frequently violated" list read on every drafting run.Improvement targets
.agents/skills/draft_docs/SKILL.md— two additive bullets in step 6.5..agents/logs/human_review_feedback.jsonl— 9 collected human-feedback records from the Step B GitHub collection (folded into this PR becausemainis branch-protected and a direct push was rejected).Patterns reviewed but not acted on
validate_ui_refsskill.aeo_crosslink_auditskill.Signal sources note
[SIGNAL:style-lint]or[SIGNAL:pr-review]markers were emitted in the past 30 days. The only drafting runs (draft_docs) targeted the archivedwarpdotdev/gitbookrepo and produced no markers and no PRs. Noreview-docs-prruns occurred.oz-for-oss[bot], etc.) were filtered out per the security boundary; only non-bot member feedback was used.Open questions for human review
Infrastructure fix (added during human review)
The open question about signal marker retrieval was investigated and resolved.
oz run get(without flags) returns only the briefRunItem.status_messagefield — not conversation content or shell stdout. Markers printed viaechoin a shell command were invisible to theimprove-drafting-skillscollector.Fix (3 files):
draft_docs/SKILL.mdstep 8 — emit[SIGNAL:style-lint]as agent text output (part of the agent's own message), not via shellecho. Agent text appears as aTextContentBlockin the conversation and is retrievable viaoz run get --conversation.review-docs-pr/SKILL.md— same fix for[SIGNAL:pr-review].improve-drafting-skills/SKILL.mdStep A — updated retrieval to useoz run get --conversationwith a precisejqcommand that recursively traverses thesteps[].messages[].content[]response structure. Also fixed an incorrect JSON path (messages[]→steps[].messages[]) and added a concretejqexample.Conversation: https://app.warp.dev/conversation/d8fc79d6-5723-42f3-a3d5-cc68a96b28a6
Run: https://oz.warp.dev/runs/019f1452-d491-7efc-95fb-d08418579969
This PR was generated with Oz.