[kyle-sexton]

Closes #13036

Summary

Adds scripts/check-view-medium-lists.mjs, run by scripts/check-changed-skills.sh on every invocation. It fails when a view lane's *.md or evals.json holds a line or JSON string naming two or more distinct backticked medium ids. Medium ids come from MEDIUMS and lanes from the lane(...) entries in lib/view-registry.mjs; CHANGELOG.md, README.md and the lib/rendered-view.md shared copies are skipped. scripts/view-medium-list-exemptions.txt exempts lanes still listing media; a row for a clean or unregistered lane fails, so the list only shrinks.

This replaces #13076, whose CI failed after its exemption list went stale.

Fix

The education, overengineering and prototype slices of #12646 (#12682, #12679, #12684) merged after #13076 was written. Their five lanes (education:illustrate, education:teach, overengineering:audit, prototype:explore-directions, prototype:pressure-test) no longer list media, so the stale-row rule failed both the current-tree test in scripts/check-view-medium-lists.test.mjs and check-skills. This PR deletes those five rows. architecture (#12672) and visualization:visualize (#12669) still list media and keep their rows; codebase-health:audit and visualization:present were already clean and carry no row.

Verification

• bash scripts/check-view-medium-lists.test.sh: 16 of 16 pass (was 15 of 16).
• bash scripts/check-changed-skills.sh origin/main: no view lane lists media (2 lane(s) exempt), exit 0.
• bash scripts/check-changed-skills.test.sh: passes.

Related

• test(rendered-views): fail a view lane's skill text that lists medium values #13076: the original lanes App pull request this replaces.
• Epic: consolidate the planning chain #12646: parent epic of the view-medium slices.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-04T06:02:24.628254Z	5b893cf	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed 5b893cf859ba8b5184748f1dd357c05fbde8859c: read 6 of 6 changed file(s); 0 finding(s), 0 as inline comments.

Security review: PR #13106, no security issues found

What I read: every hunk of all 6 changed files in .claude-lane/pr.diff, plus the PR context and the prior-findings file (0 earlier rounds, cap 3, not past the cap). None of the changed files matches a trusted-path glob. None is a trust control (trusted-trigger, lane token, kill switch, token broker, merge hold or .github/actions/*). None is a plugin manifest or a standing instruction surface.

What changed: the PR adds a repository lint gate, scripts/check-view-medium-lists.mjs, which scripts/check-changed-skills.sh calls on every run. The gate fails when a view lane's *.md or evals.json names two or more backticked medium ids on one line. The PR also adds its tests and a list of exempt lanes (scripts/view-medium-list-exemptions.txt).

Paths I checked:

• No code execution from the checked tree. The gate reads lib/view-registry.mjs as text with regexes (readRegistry) and never imports it. Only the test file imports the registry, and tests run on the repository's own tree.
• Fail-closed wiring. check-changed-skills.sh runs the checker from the script's own directory ($SCRIPT_DIR), not from the checked tree. If the checker is missing it exits 2, and any return code above 1 also exits 2. A medium-list finding (rc 1) now carries through both the "nothing to gate" path and the final exit. The gate tightens checking and loosens nothing.
• File walk. readdirSync(..., { withFileTypes: true }) with isDirectory() / isFile() does not follow symlinks, so a symlink inside a lane directory is skipped. The walk reads only .md and evals.json files, and the only output is line-number and medium-id hits on stderr and stdout. Lane ids come from the repository's own registry. A crafted id could at most point this read-only walk at other paths, and nothing is written or sent anywhere, so no leak, compromise or charge path exists.
• Exemption list. This is a data file for the lint gate. It gates no merge hold, token or trust decision.

PR context: it contains no embedded instructions aimed at the review lane.

Spend: none beyond the normal CI test time for one small extra Node check per check-skills run.

critical-findings: 0

[github-actions]

claude-review reviewed 5b893cf859ba8b5184748f1dd357c05fbde8859c: read 6 of 6 changed file(s); 0 finding(s), 0 as inline comments.

Code review: no findings

I read every hunk of all 6 changed files. To check how the checker parses its inputs, I also read lib/view-registry.mjs, the start of scripts/check-changed-skills.test.sh (its helpers), the exit paths in scripts/check-changed-skills.sh, and the lib/rendered-view.md rows in scripts/shared-copies.txt. I ran nothing. The Verification section's results (16 of 16 tests passing, check-changed-skills.sh origin/main exiting 0 with 2 lanes exempt) are the author's claim and I have not confirmed them.

How the change works

• Reading the registry as text.
  • readRegistry matches how lib/view-registry.mjs is actually written today: the MEDIUMS block opens with Object.freeze({ and a newline, each key starts its own line, and lanes appear as lane("<id>", …). The const lane = ( definition does not match \blane\(.
  • The test compares the parsed result with Object.keys(MEDIUMS) and Object.keys(LANES). That works because LANES is an object keyed by lane id, built in insertion order.
  • If the registry is refactored so the parse finds no mediums or no lanes, the check exits 2 rather than passing silently.
• Skipping shared copies. scripts/shared-copies.txt rows have the form lib/rendered-view.md plugins/<p>/reference/rendered-view.md. These match the ${dir}/${name} relative paths that walk builds, so the copies are skipped as intended.
• Wiring into check-changed-skills.sh.
  • The script runs the check for every invocation and treats any exit above 1 as an environment error (exit 2).
  • It feeds medium_lists into both the "no changed skills" exit and the final failure aggregation. No other exit 0 path skips it.
  • In --merge-base mode it runs the trusted $SCRIPT_DIR copy against --root "$PWD" and reads the checked tree only as text, the same way the evals-warrant exemptions are handled.
  • Existing test scenarios stage no lib/view-registry.mjs, so the check doesn't run in them.
• The new check-changed-skills.test.sh case. base_sha returns HEAD after the commit, so the second run really has no changed skill. The test therefore exercises the "fails though no skill changed" path its name describes.
• Exemption rows. Rows that name an unregistered lane, a clean lane, a duplicate lane, or cite no issue each produce a finding. Each case has a test.

Closing issue #13036

• Done when 1 (met). The check runs from check-changed-skills.sh, takes the medium ids from MEDIUMS and the lanes from the lane(...) entries, and scans only *.md and evals.json. The scope-out test in check-view-medium-lists.test.mjs covers the skipped files (CHANGELOG.md, README.md, shared copies, other file types and other skills).
• Done when 2 (met). The tests cover a failing list in skill markdown, in a bare lane's plugin directory and in an evals.json string. They also cover a passing single-medium fixture prompt and the current tree.
• Done when 3 (met). The exemption file holds the two lanes that still list media, each citing its open slice. The rule that a row for a clean lane fails keeps the list from growing. Whether exactly these two rows are needed rests on the "passes the current tree" test, which I did not run.

Spend: none.

[github-actions]

Post-merge mutation testing

• Range: bae01c3..38ec1a6
• Merged pull request: #13106
• Outcome: reported

Killed 5, survived 1, timeout 0, no coverage 0, invalid 0, not run 1.
Covered-code score 83%; mutation score 83%.

File	Killed	Survived	Timeout	No coverage	Invalid	Not run
lib/view-registry.mjs	0	0	0	0	0	1
lib/view-settings.mjs	1	0	0	0	0	0
scripts/check-changed-skills.sh	2	1	0	0	0	0
scripts/check-view-medium-lists.mjs	2	0	0	0	0	0

Surviving mutants

Location	Operator	Disposition	Cause
scripts/check-changed-skills.sh:284	LCR	productive	input-gap

The run's summary has the full report.