Port Ricky review-depth skill updates#63
Conversation
📝 WalkthroughWalkthroughThis PR updates the ChangesReview-Depth Selection Model Rollout
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the documentation and package metadata to transition from a rigid "mandatory Claude-then-Codex" review loop to a flexible "review-depth" model. Under this new model, deep-tier workflows still require the full sequential Claude and Codex loops, while light and standard workflows can scale down based on the selected tier, provided that deterministic gates and at least one independent Claude review pass remain. The review feedback focuses on improving grammatical consistency, tense alignment (e.g., using "selects" instead of "selected"), and phrasing clarity across the updated markdown files.
| 4. A fresh self-review agent reads the actual files, AGENTS.md / CLAUDE.md, recent related work, and local conventions. It writes findings to disk. | ||
| 5. The implementer repairs valid findings, then deterministic gates rerun from captured output. | ||
| 6. After all squads converge, run the mandatory sequential fresh-eyes review/fix loops: Claude reviews the final diff and artifacts, a fixer repairs valid findings and adds or updates appropriate tests/proofs, Claude reviews the post-fix state again, then Codex repeats the same cycle from scratch over the post-Claude-fix state. | ||
| 6. After all squads converge, run the review-depth fresh-eyes review/fix loops. Deep-tier workflows require Claude to review the final diff and artifacts, a fixer to repair valid findings and add or update appropriate tests/proofs, Claude to review the post-fix state again, then Codex to repeat the same cycle from scratch over the post-Claude-fix state. Light and standard generated workflows may scale down only when Ricky selected that tier and deterministic gates plus at least one independent Claude review/fix pass remain mandatory. |
There was a problem hiding this comment.
To maintain consistent tense with the surrounding verbs ("remain", "may scale down"), consider using "selects" or "has selected" instead of the past-tense "selected".
| 6. After all squads converge, run the review-depth fresh-eyes review/fix loops. Deep-tier workflows require Claude to review the final diff and artifacts, a fixer to repair valid findings and add or update appropriate tests/proofs, Claude to review the post-fix state again, then Codex to repeat the same cycle from scratch over the post-Claude-fix state. Light and standard generated workflows may scale down only when Ricky selected that tier and deterministic gates plus at least one independent Claude review/fix pass remain mandatory. | |
| 6. After all squads converge, run the review-depth fresh-eyes review/fix loops. Deep-tier workflows require Claude to review the final diff and artifacts, a fixer to repair valid findings and add or update appropriate tests/proofs, Claude to review the post-fix state again, then Codex to repeat the same cycle from scratch over the post-Claude-fix state. Light and standard generated workflows may scale down only when Ricky selects that tier and deterministic gates plus at least one independent Claude review/fix pass remain mandatory. |
| ## Review-Depth Fresh-Eyes Loops | ||
|
|
||
| Every workflow must include two comprehensive fresh-eyes review/fix loops before final acceptance, commit, PR creation, or handoff: first Claude, then Codex. This applies even to small workflows and even when deterministic tests pass. Tests prove commands passed; the fresh-eyes loops make independent agents read the actual resulting files and artifacts as if they did not author them. | ||
| Deep-tier workflows must include two comprehensive fresh-eyes review/fix loops before final acceptance, commit, PR creation, or handoff: first Claude, then Codex. Lighter generated workflows may use a scaled review depth only when Ricky classifies the spec as light or standard, deterministic tests and hard validation stay mandatory, and at least one independent Claude review/fix pass reads the resulting files and artifacts as if it did not author them. |
There was a problem hiding this comment.
The phrase "as if it did not author them" is slightly awkward because "it" grammatically refers to "pass" (which doesn't author files). It would be clearer to refer to the "reviewing agent" or "reviewer" instead.
| Deep-tier workflows must include two comprehensive fresh-eyes review/fix loops before final acceptance, commit, PR creation, or handoff: first Claude, then Codex. Lighter generated workflows may use a scaled review depth only when Ricky classifies the spec as light or standard, deterministic tests and hard validation stay mandatory, and at least one independent Claude review/fix pass reads the resulting files and artifacts as if it did not author them. | |
| Deep-tier workflows must include two comprehensive fresh-eyes review/fix loops before final acceptance, commit, PR creation, or handoff: first Claude, then Codex. Lighter generated workflows may use a scaled review depth only when Ricky classifies the spec as light or standard, deterministic tests and hard validation stay mandatory, and at least one independent Claude review/fix pass reads the resulting files and artifacts as if the reviewing agent did not author them. |
| 8. Run the mandatory fresh-eyes review loops in sequence: Claude reviews the actual final diff and artifacts, a fixer repairs findings and hardens them with appropriate tests/proofs, Claude reviews the post-fix state again, then Codex repeats the same cycle from scratch over the post-Claude-fix state. | ||
| 9. Optional extra reviewers can be added for high-stakes work, but they do not replace the sequential Claude-then-Codex loops. | ||
| 10. Final signoff only happens after post-Codex-fix review and final deterministic gates prove the spec is complete, or a blocker artifact explains why it cannot be completed. | ||
| 8. Run the selected review-depth fresh-eyes loops. Deep-tier workflows run the full Claude loop followed by the Codex loop; light and standard generated workflows may scale down only when Ricky selected that tier. |
There was a problem hiding this comment.
To maintain consistent tense with the surrounding verbs ("run", "may scale down"), consider using "selects" or "has selected" instead of the past-tense "selected".
| 8. Run the selected review-depth fresh-eyes loops. Deep-tier workflows run the full Claude loop followed by the Codex loop; light and standard generated workflows may scale down only when Ricky selected that tier. | |
| 8. Run the selected review-depth fresh-eyes loops. Deep-tier workflows run the full Claude loop followed by the Codex loop; light and standard generated workflows may scale down only when Ricky selects that tier. |
| | Final acceptance before repair and dual review | Broken work can stop or commit without giving the team a final chance to fix it. Run repairable gates first, then the Claude-then-Codex review/fix loops, then final deterministic acceptance before commit/PR | | ||
| | Skipping the mandatory dual review loops | Add sequential Claude-then-Codex fresh-eyes review/fix loops after repairable verification and before final acceptance, commit, PR creation, or handoff | | ||
| | Final acceptance before repair and required review | Broken work can stop or commit without giving the team a final chance to fix it. Run repairable gates first, then the selected review-depth review/fix loop, then final deterministic acceptance before commit/PR | | ||
| | Skipping required review-depth loops | Add the review/fix loop required for the generated review depth after repairable verification and before final acceptance, commit, PR creation, or handoff; deep tier requires sequential Claude-then-Codex fresh-eyes loops | |
There was a problem hiding this comment.
To maintain terminology consistency with the rest of the document (which uses "selected review-depth loops" or "selected review depth"), consider changing "generated review depth" to "selected review depth".
| | Skipping required review-depth loops | Add the review/fix loop required for the generated review depth after repairable verification and before final acceptance, commit, PR creation, or handoff; deep tier requires sequential Claude-then-Codex fresh-eyes loops | | |
| | Skipping required review-depth loops | Add the review/fix loop required for the selected review depth after repairable verification and before final acceptance, commit, PR creation, or handoff; deep tier requires sequential Claude-then-Codex fresh-eyes loops | |
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
| 2. Pick the coordination shape deliberately: Conversation for non-trivial coordination, Pipeline only for linear one-shot handoffs. | ||
| 3. Use repairable validation gates: capture red output with `failOnError: false`, hand it to a repair owner, then rerun the same check. | ||
| 4. Run the mandatory fresh-eyes loops in order: Claude review/fix/final review/final fix, then Codex review/fix/final review/final fix. | ||
| 4. Run fresh-eyes review at the depth warranted by the spec: deep-tier workflows use Claude review/fix/final review/final fix followed by Codex review/fix/final review/final fix; lighter generated workflows may scale down only when deterministic gates, hard validation, and at least one independent Claude review/fix pass remain on the critical path. |
There was a problem hiding this comment.
🟡 Incomplete terminology migration leaves contradictory "mandatory" instruction in Pipeline table
Line 32 introduces the new review-depth concept: "lighter generated workflows may scale down." Line 1580 (also changed in this PR) updates the Pipeline section to say "the selected review-depth review/fix loops." However, line 93 in the same file still says "The mandatory final Claude-then-Codex review/fix loops still apply" in the Pipeline shape table — directly contradicting the new variable-depth concept. An AI agent reading this skill will receive conflicting instructions: the checklist says review depth is variable based on spec tier, but the Pipeline table says the full loops are always mandatory.
Contradicting line at line 93 (unchanged)
skills/writing-agent-relay-workflows/SKILL.md:93:
| Pipeline (one-shot DAG) | ... | Linear, well-specified transformations; deterministic data passing; no live agent-to-agent coordination during implementation. The mandatory final Claude-then-Codex review/fix loops still apply. |
This contradicts the changed line 32 and changed line 1580 which both use "review-depth" / "selected" language.
Prompt for agents
The PR migrates terminology from 'mandatory Claude-then-Codex' to 'review-depth' across the skill file, but missed updating line 93 in the Coordination Style table. Line 93 still reads 'The mandatory final Claude-then-Codex review/fix loops still apply' in the Pipeline shape's 'Use when' column. This should be updated to use the new review-depth language (e.g., 'The selected review-depth fresh-eyes loops still apply') to match line 32 (checklist item 4) and line 1580 (Pipeline subsection heading text) which both now use the variable-depth terminology. The file is skills/writing-agent-relay-workflows/SKILL.md and the problematic text is in the markdown table at line 93.
Was this helpful? React with 👍 or 👎 to provide feedback.
|
Reviewed PR #63 and fixed the issues found. Changes made:
Local checks run:
|
|
✅ pr-reviewer applied fixes — committed and pushed Reviewed PR #63 and fixed the issues found. Changes made:
Local checks run:
|
![agent-relay-bot[bot]](https://avatars.githubusercontent.com/in/3624277?s=60&v=4){kind=small}
There was a problem hiding this comment.
✅ pr-reviewer applied fixes — committed and pushed 9bc4ff5 to this PR. The notes below describe what changed.
Reviewed PR #63 and fixed the issues found.
Changes made:
- Updated stale
/create-workflowcommand wording to the new review-depth model. - Bumped
/create-workflowmetadata from1.0.3to1.0.4inprpm.jsonandREADME.md. - Fixed remaining unconditional Codex/post-Codex acceptance language so light/standard tiers are coherent.
- Applied all bot review suggestions around tense, terminology, and ambiguous phrasing.
Local checks run:
- Parsed
prpm.jsonandprpm.lock. - Verified manifest package files exist and README versions align for changed packages.
- Scanned focused files for stale old-contract phrases.
- Checked changed files for trailing whitespace.
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
1 issue found across 4 files
Reply with feedback, questions, or to request a fix.
Re-trigger cubic
|
Reviewed PR #63 locally and fixed the docs consistency issues I found. Changes made:
Validation run:
|
|
✅ pr-reviewer applied fixes — committed and pushed Reviewed PR #63 locally and fixed the docs consistency issues I found. Changes made:
Validation run:
|
There was a problem hiding this comment.
✅ pr-reviewer applied fixes — committed and pushed 51cb9e4 to this PR. The notes below describe what changed.
Reviewed PR #63 locally and fixed the docs consistency issues I found.
Changes made:
- Renamed the stale unconditional “Mandatory Claude-Then-Codex” common-pattern section to “Deep-Tier Claude-Then-Codex”.
- Clarified that the full Claude-then-Codex loop example applies when selected review depth is deep.
- Replaced generic-skill references to “Ricky selects/classifies” with workflow-spec classification language in both affected skills.
Validation run:
jq empty prpm.json- Metadata/file/README version consistency check across all 22 package entries
- Repo-wide stale phrase scan for old mandatory/unconditional review-loop wording
1 participant
Summary
writing-agent-relay-workflowsandrelay-80-100-workflowskillslight,standard, anddeepreview/fix paths, final-review gate dependencies, and required artifacts/create-workflowwith selected review-depth behavioragent-workforce-skillsto 1.1.1,writing-agent-relay-workflowsto 1.6.16,relay-80-100-workflowto 1.0.8, andcreate-workflowto 1.0.4Validation
node -e 'JSON.parse(require("fs").readFileSync("prpm.json","utf8")); console.log("prpm.json OK")'git diff --checkrg -n "after the Codex loop|post-Codex-fix review are green|mandatory Claude|mandatory sequential|dual review loops|mandatory dual review|Every workflow must include two|This applies even to small|post-Codex-fix path|Claude-then-Codex review/fix loops|mandatory final Claude-then-Codex" skills/writing-agent-relay-workflows/SKILL.md skills/relay-80-100-workflow/SKILL.md commands/create-workflow.md README.md prpm.jsonreturned no matchesprpm publish --dry-run --package writing-agent-relay-workflowsprpm publish --dry-run --package relay-80-100-workflowprpm publish --dry-run --package create-workflowNote: full
prpm publish --dry-runpreviously reported an existing unrelated frontmatter validation issue inskills/openclaw-orchestrator/SKILL.md; scoped dry-runs for the changed packages pass.