Repository navigation
feat(docs-naming): read per-person config from a file_names userConfig option - #18067
melodic-automation-lanes[bot] wants to merge 2 commits into
Conversation
Lane-Activity: work-resolve-issue/resolve-issue
Lane run report
Run 44957495349, attempt 1
Failed GitHub API callsNo failed call recorded by the run job's scripts. |
| "file_names": { | ||
| "type": "string", | ||
| "title": "Your file-name settings (JSON)", | ||
| "description": "Your file_names keys as one JSON object, the shape of docs/conventions/docs-naming.yaml, e.g. {\"exempt_basenames\":[\"NOTES.md\"]}. Empty (default): none. Repository files win; on policy-floor keys it only adds, and generated is ignored. Invalid JSON is reported and ignored.", | ||
| "default": "" | ||
| } |
There was a problem hiding this comment.
manifest userConfig.file_names: flag, added free-form string option with no options list (not a selector, not inert). Skill text substitutes ${user_config.file_names} into single-quoted Bash arguments for setup-check.sh, setup-apply.sh, inventory.sh, sweep.sh, emit-gate.sh and apply-rename.sh, and resolve-config.sh reads it
What installed users get on update: a new per-person config layer. The default "" is unset, so nothing changes until a user sets the option. Once it is set, the value joins the file_names merge between the bundled defaults and the team layer.
What bounds it:
- It can never supply the team-only
generatedregenerator command (resolve-config.shL345-357). - Its tiers are stamped
_layer: userConfig, andsweep.shL156 treats that layer as non-authoritative, so they cannot override a team tier. - The scripts hand the value to
jqon stdin and neverevalit.
Remaining exposure: when the agent builds the Bash command, keeping the value inside one single-quoted argument depends on the model following the skill's "escape any single quote in it" instruction. A value that contains ' and is pasted without escaping would break out of the argument. The value comes from the operator's own user settings, so only the operator can hurt themselves this way, not a third party. If you want this to rely on code rather than the model, pass the value on stdin or in a file instead of in the command line.
severity: IMPORTANT
claude-lane: claude-security-review
There was a problem hiding this comment.
pr-unblock / address-feedback: UNCERTAIN, left open
Evidence: plugins/docs-naming/.claude-plugin/plugin.json:29, plugins/docs-naming/skills/setup/SKILL.md:31, plugins/docs-naming/scripts/resolve-config.sh:345
The code confirms what the finding describes. The file_names option is a free-form string. Four skills tell the agent to pass it as one single-quoted --user-config argument and to escape any single quote in it (setup/SKILL.md:31 and the matching lines in the audit, gate and realign skills). Once inside the scripts, the value goes to jq as data and is never evaluated. The resolver also keeps it away from the team-only generated command (resolve-config.sh:345). So the only gap left is the model's quoting, and only the operator's own setting can trigger it. The suggested fix would pass the value on stdin or in a file instead. That changes how six scripts (setup-check, setup-apply, inventory, sweep, emit-gate, apply-rename) take their input, plus the four skill bodies and their allowed-tools patterns. That is too big for one scoped change in this round, and it is a security design choice that the code alone cannot decide. A person should choose between keeping the quoting instruction, given the self-only exposure, and moving the value to stdin or a file in a follow-up.
|
claude-security-review reviewed Security review: PR #18067, changes since 50ae432 (blocking only)This is an incremental, blocking-only review: this lane already reviewed an earlier head, so it reports only CRITICAL and IMPORTANT findings. I read every hunk of the 6 files in scope. The new push adds no security finding. The one finding counted below is the earlier manifest flag, which is still open and whose lines have not changed. What I checked
Prior finding. The critical-findings: 0 Manifest lines (1)
|
|
claude-review reviewed CI code review: PR #18067, incremental (50ae432 → 082571e)Blocking only: this lane already reviewed an earlier head, so this pass reports blocking findings only. I read all six files in the incremental scope, every hunk of each, and found none. Both earlier findings are fixed in this push:
Loosened checks. Not verified: this lane has no shell, so neither Verifier dropped 0 of 0 candidates. |
Lane run report
Run 44958804226, attempt 1
Failed GitHub API callsNo failed call recorded by the run job's scripts. |
Lane-Activity: pr-unblock/address-feedback Lane-Thread: PRRT_kwDOTCGFQM66dTLJ Lane-Thread: PRRT_kwDOTCGFQM66dTLO
Lane run report
Run 44958804226, attempt 1
Failed GitHub API callsNo failed call recorded by the run job's scripts. |
|
pr-unblock / address-feedback finished round 1 of max 4 on 082571e. |
Lane run report
Run 44958804226, attempt 1
Failed GitHub API callsNo failed call recorded by the run job's scripts. |
|
The pr-unblock lane stopped
When a retry on this head can change the outcome, the lane runs one on its own, up to its per-head cap, and edits this comment if it stops again; a run that does not stop clears the status. When no retry can (a review thread the lane does not take or settle, or a failing check after the retries are spent), a person resolves the threads or gets the checks green, then dispatches the workflow below: that run finds the cause gone and clears the status. On a pull request the lanes App opened, the dispatchers start that run once per stop; on any other, a person does. A new commit to the branch starts clean. To run it again on this head now, dispatch the gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=18067 -f activity=address-feedback |
Lane run report
Run 44958804226, attempt 1
Failed GitHub API callsNo failed call recorded by the run job's scripts. |
|
Unattended merge: native auto-merge leaves this pull request unarmed at head |
Closes #18002
Summary
Opened by the work-resolve-issue lane. The run's own summary:
The run went past its turn budget (turns=231 max-turns=225): the work may be too big for one run.
Not verified
Fix
Verification
The runner ran each check from the base copy against 50ae432:
The run reports:
Left out: the items under Not verified.
Related
#18002