docs(skills): path anchoring, promised templates shipped, neutral framing, central routing doctrine (M4)#11
Merged
Merged
Conversation
…tem 2)
loop-contract's scaffold tree + repo-os-contract/platform-map/README all
promise scripts/{verify-safety,judge-rubric,extract-trace-metrics} but
templates/ shipped only verify-fast.sh + verify-full.sh. Add the 3 as
minimal, runnable-stub .sh templates consistent with the existing two, make
verify-full.sh's commented gate examples reference the shipped names, and
clarify in loop-contract which two auto-scaffold vs which three are copy-in
stubs. Sync structural.json template_filenames (11 -> 14).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…KILLS item 3) The scaffolded WORKFLOW.md and the loop-architect realization list named /verify-slice (claude-code-orchestration) as THE acceptance-verification path, making the author's private stack look required. Reframe both so the bundled verify-fast->verify-full gate is the default and /verify-slice is a clearly optional layer on the same exit codes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…SKILLS item 4) The read->haiku / reason->sonnet / write->opus table + rationale were repeated across 7 SKILL.md files. Add reference/model-routing.md as the single source of truth (canonical tier table, rationale, optional author-stack enforcement) and trim each spoke to a self-sufficient one-line inline rule + a pointer. Sync structural.json reference_filenames (7 -> 8); references-used stays green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…S item 1)
Bare relative pointers like reference/safety-and-approvals.md break when a skill
is invoked from another CWD: they resolve against skills/<name>/ (where the file
is not) instead of the plugin root (where the shared reference/ ships). Add a
per-skill "Base directory" note anchoring reference/, templates/, schemas/, and
the bundled scripts/*.py tools to ${CLAUDE_PLUGIN_ROOT} (../../ from the skill
folder), while calling out that the co-located reference/patterns.md and the
operated loop's own scripts/verify-* + .loop/ are NOT plugin-root. The two
referenced-as-missing files (safety-and-approvals.md, prompt-templates.md) were
never missing — they ship at reference/; the pointers were just mis-anchored.
Cross-reference sweep: 115 bundled pointers resolve, 0 unresolved.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
There was a problem hiding this comment.
Pull request overview
Polishes the loop-engineer skill docs and scaffolding templates to clarify plugin-root-relative paths, ship promised verification template stubs, and centralize the model-routing doctrine into a single canonical reference.
Changes:
- Added new bundled template stubs (
verify-safety.sh,judge-rubric.sh,extract-trace-metrics.sh) and updated verification docs to treatverify-*scripts as the default acceptance gate (with/verify-sliceexplicitly optional). - Centralized the model-routing rule into
reference/model-routing.mdand updated multiple skills to point to it. - Updated structural eval case lists to enumerate the new reference/template filenames.
Reviewed changes
Copilot reviewed 18 out of 18 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| templates/WORKFLOW.md.tmpl | Clarifies default acceptance verification as bundled verify-* scripts; makes /verify-slice explicitly optional. |
| templates/verify-safety.sh | Adds a safety/approval/injection gate stub with an initial “no hardcoded secrets” check. |
| templates/verify-full.sh | Updates verification suite template comments to include advisory judging + safety/regression gating examples. |
| templates/judge-rubric.sh | Adds an advisory-only rubric judging harness stub that always exits 0. |
| templates/extract-trace-metrics.sh | Adds a metrics extraction stub emitting a minimal JSON skeleton. |
| skills/loop-runtime-monitor/SKILL.md | Adds plugin-root base-directory anchoring and points model-routing guidance to the canonical reference. |
| skills/loop-runtime-monitor/reference/patterns.md | Adds plugin-root base-directory anchoring for referenced scripts. |
| skills/loop-run/SKILL.md | Adds plugin-root base-directory anchoring and centralizes model-routing details via the new reference doc. |
| skills/loop-repair/SKILL.md | Adds plugin-root base-directory anchoring and points model-routing guidance to canonical reference. |
| skills/loop-inspector/SKILL.md | Adds plugin-root base-directory anchoring and clarifies which filenames are signals vs bundled files. |
| skills/loop-inspector/reference/patterns.md | Adds plugin-root base-directory anchoring and clarifies signals vs bundled files. |
| skills/loop-flywheel/SKILL.md | Adds plugin-root base-directory anchoring and points model-routing to canonical reference. |
| skills/loop-evals/SKILL.md | Adds plugin-root base-directory anchoring and points model-routing to canonical reference. |
| skills/loop-engineer/SKILL.md | Adds plugin-root base-directory anchoring and adds canonical reference/model-routing.md to the router’s references list. |
| skills/loop-contract/SKILL.md | Adds plugin-root base-directory anchoring and clarifies which verification templates scaffold vs copy-in from templates/. |
| skills/loop-architect/SKILL.md | Adds plugin-root base-directory anchoring, clarifies default acceptance gating, and points model-routing to canonical reference. |
| reference/model-routing.md | New canonical model-routing doctrine (tier table + rationale + optional enforcement notes). |
| evals/cases/structural.json | Additively enumerates new reference/template filenames for structural self-eval checks. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| # --- GATE 4: artifact quality check --- | ||
| # e.g. python3 scripts/judge-rubric.py --artifacts artifacts/ --rubric EVALS/rubrics/main.md | ||
| # --- GATE 4: artifact quality check (advisory — never blocks) --- | ||
| # e.g. bash scripts/judge-rubric "$WORKSPACE" EVALS/rubrics/main.md |
| # --- GATE 6: regression suite --- | ||
| # e.g. python3 scripts/extract-trace-metrics.py EVALS/traces/ | python3 scripts/verify-safety.py | ||
| # --- GATE 6: regression + safety suite --- | ||
| # e.g. bash scripts/extract-trace-metrics "$WORKSPACE" && bash scripts/verify-safety "$WORKSPACE" |
Comment on lines
+19
to
+23
| if grep -rInE '(api[_-]?key|secret|passwd|password|token)[[:space:]]*=[[:space:]]*[^[:space:]]' \ | ||
| --exclude-dir=.git --exclude-dir=.loop "$WORKSPACE"/src 2>/dev/null; then | ||
| echo "[verify-safety] FAIL: assigned secret literal found — move it to the environment" | ||
| exit 1 | ||
| fi |
Comment on lines
+17
to
+18
| # Flags a label immediately assigned an inline value (env-var NAMES alone are fine). | ||
| # Tighten the value side to a quoted 8+ char token once this loop has real source. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8882d2ed12
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+19
to
+20
| if grep -rInE '(api[_-]?key|secret|passwd|password|token)[[:space:]]*=[[:space:]]*[^[:space:]]' \ | ||
| --exclude-dir=.git --exclude-dir=.loop "$WORKSPACE"/src 2>/dev/null; then |
There was a problem hiding this comment.
For workspaces whose code or contract files live outside src/ (for example docs-only loops, app/, lib/, root scripts, or WORKFLOW.md), this blocking safety gate exits PASS because grep is run only over $WORKSPACE/src and missing-directory errors are suppressed. That means an inline api_key = ... in the tracked contract or non-src source is missed despite this template advertising a real secret-literal safety check; scan tracked files/the workspace with excludes instead of only src.
Useful? React with 👍 / 👎.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
M4 skills polish batch (launch criterion S6):
reference/,templates/,schemas/, and bundled scripts to the plugin root; the loop-run reference pointers (safety-and-approvals.md,prompt-templates.md) were mis-anchored to the skill dir — the files ship at plugin-rootreference/and now resolve. Cross-reference sweep: 115 bundled pointers resolved, 0 unresolved.verify-safety.sh,judge-rubric.sh,extract-trace-metrics.shadded as minimal runnable stubs consistent with the existing pair; loop-contract now says honestly which templates auto-scaffold vs copy-in./verify-slicereferences now name the bundledverify-fast→verify-fullgate as the default, with/verify-sliceas a clearly-optional integration.reference/model-routing.md; 7 SKILL.md files trimmed to a one-line rule + pointer.evals/cases/structural.jsonupdated additively only (new template/reference filenames enumerated per the file's own instruction — no check logic altered).Gates: self_eval 13/13, frontmatter 9/9 clean, suite 118+2s / 120 (jsonschema),
claude plugin validate --strictpassed.