[melodic-automation-lanes]

Closes #13080

Summary

Opened by the work-resolve-issue lane. The run's own summary:

Each lane registration in lib/view-registry.mjs now carries a hasMediumOption flag, and lib/view-settings.mjs appends '; per user, the <plugin> plugin's medium option also works' to the retired ~/.claude/rendered-views.md notice only when the lane's flag is set. The flag values come from each plugin's plugin.json, and a new test enforces that they match. Education does declare a medium option, so its two lanes are true, which corrects the intake refinement's claim. Architecture, overengineering, prototype (both lanes) and ai-briefing are false. The plugin copies were regenerated with sync-shared-copies. Overengineering eval 15 no longer expects the clause. The prototype and ai-briefing skill docs no longer tell the agent to drop or rewrite the notice by hand: they print it as the resolver gives it.

Not verified

• lib/view-settings.test.mjs (node --test) was not executed: this lane allows no interpreter, so the new tests are left to the PR's CI
• tidy skipped: the change is 39 tracked files plus 16 fragments, over the 10-file and 300-line cutoffs, and touches the shared path lib/
• The intake refinement listed education:illustrate and education:teach as having no medium option; plugins/education/.claude-plugin/plugin.json declares one, so both lanes are set true
• check-affected reported typos as not installed and left several whole-repo scanners to CI

Fix

• lib/view-registry.mjs: lane() takes a fifth hasMediumOption argument, every LANES entry sets it explicitly, and the header comment documents it
• lib/view-settings.mjs: retiredNotices takes hasMediumOption and adds the per-user clause only for a claude-user notice on a lane with the flag; resolveMedium passes laneEntry.hasMediumOption
• lib/view-settings.test.mjs: new test asserting the planning lane's notice ends with the per-user clause and the overengineering:audit lane's notice has no 'per user'; new test asserting every lane's hasMediumOption matches whether plugins/<plugin>/.claude-plugin/plugin.json declares…
• plugins/{review,visualization,work-items,harness-ops,source-control,planning,adhd,ai-briefing,architecture,debugging,discovery,education,prototype,codebase-health,overengineering}/lib/view-settings.mjs: regenerated copies
• plugins/{review,visualization,work-items,harness-ops,source-control,planning,machine-health,adhd,ai-briefing,architecture,debugging,discovery,education,prototype,codebase-health,overengineering}/lib/view-registry.mjs: regenerated copies
• plugins/overengineering/skills/audit/evals/evals.json: eval 15 fixture notice drops the per-user clause (no expectation asserted the clause)
• plugins/prototype/skills/explore-directions/reference/medium.md: notices are printed as they stand; the instruction to drop the clause by hand is removed
• plugins/prototype/skills/pressure-test/SKILL.md: same as medium.md
• plugins/ai-briefing/skills/generate/SKILL.md: removed the hand-rewrite exception for the retired user-file notice, which no longer matched the resolver's text; notices print as they stand
• plugins/ai-briefing/skills/generate/evals/evals.json: eval 11 expects the resolver's own caveat '; it publishes only while that file is untracked and gitignored' in place of the hand-appended one
• .changes/<plugin>/work-resolve-issue-issue-13080-41774896462-1-*.md: patch fragments for all 16 carrier plugins (Fixed for architecture, overengineering, prototype, ai-briefing; Changed for the rest)

Verification

The runner ran each check from the base copy against 7b9d914:

• scripts/check-changelog-fragments.sh --check: passed
• scripts/check-changed-skills.sh --merge-base: passed

The run reports:

• grep of plugins/*/.claude-plugin/plugin.json for a medium userConfig key: planning, work-items, debugging, education, adhd, visualization, codebase-health, discovery and review declare one; architecture, overengineering, prototype and ai-briefing do not
• .lane/verify/bin/sync-shared-copies: regenerated 31 copies, exit 0
• .lane/checks/bin/check-changelog-fragments: all 22 fragments valid (run before and after the review fix)
• .lane/checks/bin/check-changed-skills: 4 skills checked, 0 failed (warnings only, all already present)
• .lane/checks/bin/check-affected: 12 passed, 7 skipped, 0 failed over 55 changed files
• .lane/verify/bin/check-evals-quality on the overengineering audit and ai-briefing generate evals: PASS, 0 warnings
• Fresh-context review subagent: no correctness findings; it derived the resolver's exact ai-briefing eval 11 notice text and confirmed it matches; one long-line style note, fixed
• The lane run: https://github.com/melodic-software/claude-code-plugins/actions/runs/41774896462

Left out: the items under Not verified.

Related

#13080

[github-actions]

claude-review reviewed 7b9d914ddf5d018a255822e74e451cf52242c5dc: read 55 of 55 changed file(s); 0 finding(s), 0 as inline comments.

Code review: PR #13104, no findings

The PR fixes #13080. Each lane in lib/view-registry.mjs now has a hasMediumOption flag. retiredNotices in lib/view-settings.mjs adds the "; per user, the plugin's medium option also works" clause only to a claude-user notice on a lane where that flag is set.

What I checked

• Resolver logic (lib/view-settings.mjs:216, :246): resolveMedium passes the flag from laneEntry. It runs only after the unknown-lane guard, so the flag is always present. The other notice kinds (claude-project, claude-local, fenced-block) are unchanged.
• Registry: every lane(...) entry sets the fifth argument explicitly. The architecture, overengineering, prototype (both lanes) and ai-briefing lanes are false; the rest are true.
• Tests (lib/view-settings.test.mjs:227-238):
  • The new notice test checks both sides: planning keeps the clause, and overengineering:audit does not.
  • resolve defaults dataDir: null, so no seen-notice state suppresses the second call.
  • Existing tests already resolve a user-home retired file without a projectRoot.
  • The manifest-sync test gets the plugin name from laneId.split(":")[0], which also works for lanes without a colon (planning, architecture). It fails if a flag drifts from plugin.json.
  • I did not run the tests. The PR author also says they were not run (no interpreter in their lane) and left them to CI.
• ai-briefing eval 11: I traced the resolver for medium: artifact in ~/.claude/rendered-views.md. It gives "…is no longer read (it sets medium: artifact); set it in docs/conventions/rendered-views.local.yaml instead; it publishes only while that file is untracked and gitignored", with no per-user clause. That matches the updated expectation. The removed SKILL.md exception quoted notice text the resolver no longer produces, so removing it is correct.
• Stale instructions: no plugin skill doc still tells the agent to drop or rewrite the clause by hand. The prototype medium.md / pressure-test/SKILL.md and ai-briefing generate/SKILL.md all now print notices as they stand. The only .md match for "per user, the" outside the changed files is in plugins/testing/skills/run-e2e/context/e2e-config.md, which is about testing config, not the rendered-views notice.
• Copies: I read the first file of each identical-copy group (plugins/adhd/lib/view-registry.mjs, plugins/adhd/lib/view-settings.mjs, and the adhd changelog fragment). They match the canonical lib/ change. Every copy path is an ordinary library or fragment path, so none of them behaves differently at its location.

Closing issue #13080, "Done when"

• "A lane whose plugin has no medium option prints the retired-file notice without the per user clause; a lane whose plugin has one still prints it": met, and covered by the new notice test plus the manifest-sync test.
• "Eval 15 in the overengineering audit no longer expects the clause": met. The fixture prompt no longer has the clause, and no remaining expectation asserts it.

Manifests: no plugin.json or marketplace.json changed. All 16 changelog fragments are bump: patch. That fits a notice-text fix (Fixed for the four plugins without the option, Changed for the rest).

Embedded text: none in the PR context that tries to direct this review.

[github-actions]

claude-security-review reviewed 7b9d914ddf5d018a255822e74e451cf52242c5dc: read 55 of 55 changed file(s); 0 finding(s), 0 as inline comments.

Security review: PR #13104

No security issues found.

What changed. Each lane in lib/view-registry.mjs now has a hasMediumOption flag. retiredNotices in lib/view-settings.mjs adds the "; per user, the plugin's medium option also works" clause only when that flag is set. The rest of the PR is:

• regenerated plugin copies (plugins/*/lib/view-registry.mjs, plugins/*/lib/view-settings.mjs);
• two new tests;
• eval fixture text in overengineering and ai-briefing;
• three skill docs that no longer rewrite notices by hand;
• 16 changelog fragments.

Paths checked

• The flag doesn't touch publish decisions. It only changes notice text (lib/view-settings.mjs#L216). These are unchanged: medium selection, fallBack, decidedBy, and the publish gate gateExplicit (L232). The flag cannot make a view publish when it would not have before.
• Removed instructions (instruction-surface lens).
  • plugins/ai-briefing/skills/generate/SKILL.md drops a hand-written step that appended a "publishing value applies only while untracked and gitignored" caveat. The resolver already emits the same caveat for a publishing value aimed at the personal file (lib/view-settings.mjs#L207-L214). The skill text was also never what enforced the rule: the resolver ignores a publishing value in a committed personal override. Nothing protective is lost.
  • The prototype medium.md and pressure-test/SKILL.md edits only stop the agent from deleting a clause from the notice text. They forbid nothing new and allow nothing new.
  • I did not open the instruction exception register file; I judged these removals by their consequences.
• No plugin manifest changes. No .claude-plugin/plugin.json or marketplace.json is in the diff, so there are no manifest lines. The new test reads plugin.json files with a path built from fixed LANES ids. That is test-only and takes no outside input.
• Generated copies. Their paths are each plugin's own lib/ directory, the same place as the base copies, so the same code does the same thing at each path. None is a hook, workflow, settings or manifest path.
• Trusted paths. None of the changed files matches a trusted-path glob. Nothing changes a trust control, token scope, kill switch or merge hold.
• Embedded text. The PR description and context contain no instructions aimed at this lane.

Spend. Nothing in this change bills or uses credits.

critical-findings: 0