docs(rules): drop review caps and align skills with current model guidance - #164
Conversation
…dance Remove the review-round and fix-commit caps from the shared rules and slopguard; agents read them as stop signals. Replace the budget eval with one that closes review on evidence. State user-over-skill precedence, delegation scope, and context-growth behavior once in the rules, and update the cross-model guidance for Opus 5.5 and GPT-6 prompting docs.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
A couple of doc updates introduce clarity/grammar issues (notably around slopguard exit-code semantics and a hard-to-parse sentence) that should be corrected to avoid operational misunderstanding.
Get a fresh assessment by requesting another Copilot review.
Review effort: Lite
Findings: 2
Open (2)
What changed in this PR
This PR updates the repository’s agent/skill documentation and slopguard guidance to remove “review round / fix commit” caps that were being misinterpreted as stop conditions, and to align wording with current multi-model guidance (including moving detailed setup/migration instructions into reference docs).
Changes:
- Removes review-budget/fix-round cap language and replaces the slopguard eval scenario with an evidence-based “close-after-fixes” eval.
- Clarifies instruction hierarchy and delegation guidance in
rules/agents.md, including explicitly stating that user instructions outrank skill/guide text. - Refactors skill docs for progressive disclosure (moving installation/migration details into references) and updates some harness documentation links/phrasing.
| File | Description |
|---|---|
| skills/slopskills/SKILL.md | Moves detailed installation/migration guidance into a dedicated reference doc and links to it. |
| skills/slopskills/references/installation.md | Adds a new reference page with detailed harness path/lockfile/global-to-local migration guidance. |
| skills/slopscriber/references/agent-first.md | Updates guidance on skill discoverability/progressive disclosure and adds harness load-limit notes + updated Codex link. |
| skills/slopprep/references/agent-guidance.md | Expands cross-model contract guidance to discourage inherited caps and pressure-wording; adds model-doc links. |
| skills/slopnanny/SKILL.md | Tightens “latest commit” settlement guidance and renames a section to “Settle”. |
| skills/slopmachine/SKILL.md | Minor wording updates around delegation and milestone/handoff reporting. |
| skills/slopguard/SKILL.md | Removes budget language, clarifies invocation intent, and updates convergence/closeout wording. |
| skills/slopguard/evals/review-budget-boundary/task.md | Removes the budget-boundary eval task. |
| skills/slopguard/evals/review-budget-boundary/criteria.json | Removes the budget-boundary eval criteria. |
| skills/slopguard/evals/close-after-fixes/task.md | Adds a new eval task focused on evidence-based closeout after fixes. |
| skills/slopguard/evals/close-after-fixes/criteria.json | Adds new eval criteria to validate the evidence-based closeout behavior. |
| skills/slopguard/agents/openai.yaml | Updates the short description to remove “one” review-round implication. |
| rules/agents.md | Removes budget caps and adds explicit instruction hierarchy + updated delegation/context-growth guidance. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 311365018a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Problem
The shared rules and slopguard capped review rounds and fix commits. Agents treated the caps as stop signals and abandoned authorized work. Several rules also predated current Claude (Opus 5.5, Fable 5.1) and GPT-6 prompting guidance.
Solution
rules/agents.mdand slopguard. Replace thereview-budget-boundaryeval withclose-after-fixes, which ends review on evidence: fixed-and-verified or rejected findings, one refreshed review after a contract-changing fix, none for clerical edits or unrequested work.rules/agents.md: user instructions outrank skill text; delegate substantial work and do small lookups directly; handle stalled delegates; context growth is not a reason to stop.slopprep) and agent-first docs (slopscriber): plain reasons instead of pressure words, unfiltered reviewer findings, no inherited caps or countdowns, harness load limits, current Codex doc URLs.references/installation.md; safety constraints stay inSKILL.md.Consumers rendered from
rules/agents.md(dotfilesAGENTS.mdtemplate) need a manual resync.