Audit file
File skill-audit findings as a GitHub issue at the plugin repo that owns the offending skill. Invoked by skill-audit Phase 7 with the target repo confirmed at its destination gate. TRIGGER when: skill-audit Phase 7 completes and user opts to file upstream. DO NOT TRIGGER when: no audit findings exist, or user wants to review findings before filing.From its SKILL.md
npx -y skills add Dev10x-Guru/Dev10x-Claude --skill audit-fileAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
11.7 KB, as published. Nobody here has run it
Audit Report — File Findings Upstream
Generate a structured GitHub issue from skill-audit findings and file it at the plugin repo that owns the offending skill.
When to Use
- Delegated by
Dev10x:skill-auditPhase 7 after the user approves upstream reporting and confirms the destination tracker at the Phase 7 sub-step B2 gate - Can also be invoked standalone with a findings file
Arguments
Two arguments:
--repo <owner>/<repo>— the target issue tracker (GH-816). Required when the caller resolved a destination. Skills from every installed plugin live under~/.claude/plugins/, so the owner cannot be inferred here — the caller resolves it viamcp__plugin_Dev10x_cli__resolve_plugin_originand confirms it with the user.- Path to a findings markdown file produced by
Dev10x:skill-audit.
Target repo resolution:
| Input | Behavior |
|---|---|
--repo passed | File there. Never override it. |
No --repo, findings file has a **Target repo**: line | Use that value. |
| Neither present | REQUIRED: Call AskUserQuestion asking which repo receives the issue (do NOT use plain text, do NOT silently default to Dev10x-Guru/dev10x-claude). Offer the detected plugin repos when the findings name one, plus a free-text override. |
Store the resolved value as $TARGET_REPO — every later step
(version detection, label sync, filing) uses it.
The findings file contains:
## Session Context
- **Repo**: {repo-name}
- **Branch**: {branch-name}
- **Date**: {audit-date}
- **Target repo**: {owner}/{repo}
## Upstream Findings
| # | Phase | Classification | Skill | Description |
|---|-------|---------------|-------|-------------|
| 1 | ... | ... | ... | ... |
## Proposed Fixes
{Grouped by skill}
If no argument is provided, check for the most recent file in
/tmp/Dev10x/skill-audit/ matching findings*.md.
Workflow
Step 1: Read findings
Read the findings file passed as argument. Validate it contains at least one finding row in the table.
If empty or missing, inform the user and exit.
Step 2: Determine plugin version
Report the version of the plugin the findings are about, not
the Dev10x version. The caller's origin resolution supplies the
marketplace, plugin, and version; when the findings file carries a
**Plugin**: {marketplace}/{plugin} {version} line, use it
directly.
Otherwise list the owning plugin's cache directory:
ls ~/.claude/plugins/cache/{marketplace}/{plugin}/
Use the version directory name (e.g., 0.19.0.dev0). If the
cache directory is not found, use unknown.
Step 3: Fictionalize proprietary information (REQUIRED)
Treat the source session as private by default. The upstream
issue is a public artifact at $TARGET_REPO and MUST
NOT disclose any identifier from a non-public repository, project,
branch, ticket tracker, file path, person, or service that is not
part of the public Dev10x plugin.
Instead of redacting private identifiers with bracketed
placeholders (<private-repo>, TICKET-NN), replace them with
similar-sounding fictional counterparts drawn from pop culture —
movies, books, TV, cartoons, video games. The resulting issue
reads as a coherent story while leaking nothing. Real precedents:
issue #68 used initech/initech-pos + reviewer skywalker;
issue #98 used Tyrell Corp + Aperture Labs.
Apply the vibe-matching guide, consistency rule, and 6-step
algorithm in
references/privacy-scrub.md to
the verbatim findings text before assembling the issue body.
REQUIRED: Call AskUserQuestion when a finding is
fundamentally about a private codebase pattern that cannot be
retold through fictional stand-ins without losing the technical
point (do NOT use plain text). Options:
- Fictionalize aggressively and file (Recommended) — pick the closest pop-culture stand-in even at the cost of some specificity
- Skip this finding — exclude from upstream, keep in local notes
Never auto-include unfictionalized text. Re-read the assembled body and verify every named entity is fictional (pop-culture sourced, not a real company / person / product) before continuing to Step 4.
Step 4: Generate issue body
Build the issue body from the fictionalized findings:
## Audit Findings
**Plugin**: {plugin} {version} (marketplace: {marketplace})
**Session context**: {fictional-org}/{fictional-repo} / {fictional-branch}
**Audit date**: {date}
### Findings
| # | Phase | Classification | Skill | Description |
|---|-------|---------------|-------|-------------|
{fictionalized rows}
### Proposed Fixes
{fictionalized fixes, grouped by skill}
### Evidence
{fictionalized transcript excerpts — 2-3 lines per finding, no
real file paths, real handles, or real product names}
The "Session context" line uses fictional stand-ins (e.g.,
initech/initech-pos / skywalker/CORE-401/death-star-3/...) so
the report reads as a coherent narrative — the upstream
maintainer does not need the real repo or branch to act on a
Dev10x plugin finding.
Step 5: Derive issue title
Use the primary skill name (most findings) as the title anchor:
- Single skill:
skill-audit findings: Dev10x:{skill} - Multiple skills:
skill-audit findings: Dev10x:{skill} (+N)
Step 6: Write body to temp file
/tmp/Dev10x/bin/mktmp.sh skill-audit upstream-issue .md
Write the assembled body to that file using the Write tool.
Step 7: Derive bundling labels (REQUIRED)
Audit issues are batched by an automated session, so similar
findings should cluster under a shared label and be discoverable
as a bundle during implementation. Derive the label set from the
fictionalized findings BEFORE filing — never hardcode a single
enhancement label.
Apply the resolution algorithm in
references/labels.md:
- Start with
enhancement - Add the audit-session bundle:
audit-YYYY-MM-DD(parse from the findings file's "Audit date" line) - Add one
skill:<name>per unique skill referenced in the findings table (strip theDev10x:prefix; fictionalized names only — see Step 3) - Scan finding descriptions + proposed fixes against the topical
heuristic table in
references/labels.md§ 4; add each matching topical label once - De-duplicate and cap at 8 labels per issue
Collect the result as a comma-separated string (e.g.,
enhancement,audit-2026-05-16,skill:work-on,routing-bypass).
Store it as $LABELS for Step 8.
Step 8: Ensure labels exist on the repo (best-effort)
GitHub fails issue creation if any label is missing, so try to reconcile the set first. Fetch the current labels once and create only the missing ones:
gh label list --repo "$TARGET_REPO" --limit 200 \
--json name -q '.[].name' > /tmp/Dev10x/skill-audit/existing-labels.txt
for label in $(echo "$LABELS" | tr ',' ' '); do
grep -qxF "$label" /tmp/Dev10x/skill-audit/existing-labels.txt || \
gh label create "$label" --repo "$TARGET_REPO" \
--color "$COLOR_FOR_CATEGORY" \
--description "$DESC_FOR_CATEGORY"
done
Colors and descriptions per category live in
references/labels.md. gh label create is idempotent here
because the loop only runs for missing labels.
Labels are best-effort — never a filing precondition (GH-931
finding 6). Every step above can fail for a reporter who is not
a maintainer of $TARGET_REPO, and filing the findings matters
more than bundling them:
| Failure | Cause | Response |
|---|---|---|
unknown command "label" for "gh" | gh build without the subcommand | Skip reconciliation entirely; go to the no-label fallback |
403 / 404 on label create | Label creation needs push access | Drop the labels that could not be created |
403 / 404 applying labels at filing | Applying labels needs push (triage) access | Re-file with no --label flag |
No-label fallback: when the label set cannot be reconciled or applied, file with no labels and record the intended set in the issue body instead, so a maintainer can apply them afterwards:
**Suggested labels** (filed without them — no push access):
`enhancement`, `audit-2026-07-31`, `skill:git`
Never abort the filing because labels are unavailable. An audit finding that never reaches the tracker is a total loss; an unlabelled one is merely harder to bundle.
Step 9: File the issue
Delegate to Dev10x:ticket-create — never use raw gh issue create.
Write the title as the first line of the temp file (followed by a
blank line and the body) to avoid permission friction from special
characters in the args string. Pass the comma-separated label set
derived in Step 7 only when Step 8 confirmed every label exists
and is applicable:
Skill(skill="Dev10x:ticket-create",
args="--repo {TARGET_REPO} --body-file {temp-file-path} --label {LABELS}")
When Step 8 hit any failure above, omit the --label flag and
rely on the in-body "Suggested labels" line instead:
Skill(skill="Dev10x:ticket-create",
args="--repo {TARGET_REPO} --body-file {temp-file-path}")
If a filing attempt with labels is rejected for permissions, retry once without the flag rather than reporting failure.
The ticket-create skill reads the first line as the title when
no --title flag is provided.
Step 10: Report result
Display the created issue URL. If filing fails, show the error and the temp file path so the user can file manually.
Important Rules
- Always use
--body-file: Never pass the body inline via--body— markdown tables break shell quoting. - Plugin skills only: This skill files issues about skills shipped by an installed plugin. User-local findings should never appear in the issue body.
- Destination is supplied, never assumed (GH-816): file at
$TARGET_REPO. A finding about a non-Dev10x plugin's skill belongs at that plugin's tracker. Do NOT fall back toDev10x-Guru/dev10x-claudewhen--repois absent — raise theAskUserQuestiongate documented under Arguments instead. - No transcript dumps: Evidence sections include 2-3 lines of context per finding, not raw transcript blocks.
- One issue per audit: Batch all findings into a single issue per audit session to avoid issue spam.
- Bundle via labels, not separate issues: Apply the label
taxonomy in
references/labels.mdso similar issues can be filtered and worked together during implementation. Never file with only the defaultenhancementlabel. - Privacy by default (Step 3): The source session may belong
to a private codebase. Fictionalize repo names, owners,
branches, tracker IDs, file paths, hostnames, and personal
identifiers using pop-culture stand-ins (movies, books,
cartoons, games) before assembling the issue body. Never use
real company, product, or person names — even ones that sound
fictional. Only the public Dev10x plugin context — skill names,
plugin file paths, and
Dev10x-Guru/Dev10x-Claudeissue/PR numbers — is allowed verbatim. Re-verify every named entity is fictional before filing.