Candid Chrome QA Fix

Take the structured findings JSON from /candid-chrome-qa and turn it into actual code fixes. Pick the findings you want to address, choose how to ship them, and let Candid handle the rest — including filing tracker issues if you want.

Quick Start

/candid-chrome-qa-fix

Resolves the latest findings file in .context/findings/, asks you which findings to fix (multi-select), and asks how to ship them.

/candid-chrome-qa-fix --severity P0,P1                          # Only urgent + high severity
/candid-chrome-qa-fix --strategy issues-only                    # File Linear issues, no fixes
/candid-chrome-qa-fix --strategy per-finding --print-links      # Emit conductor:// deep links (dry run)
/candid-chrome-qa-fix --strategy batched --create-issues --tracker-team ENG

How It Works

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Candid Chrome QA Fix
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Findings file: .context/findings/2026-04-25-agent-config.json

Steps:
  1. ✅ Resolve findings file (latest, --file, or pick from recent)
  2. 🔍 Validate v2.0 schema, drop coverage receipts
  3. ☑️  Multi-select which findings to fix
  4. 🚀 Pick a strategy (batched / per-finding / local / issues-only)
  5. 🎯 Optionally file tracker issues (Linear today)
  6. 🛠️  Apply fixes — sequentially or via Conductor deep links
  7. 📋 Open PR(s) and link issues
  8. 📊 Final summary

Issue creation runs before any code change so issue IDs flow into commits, PR titles, PR bodies, and per-finding spawned-workspace prompts.

Selection UI

The skill’s selector adapts to the number of fixable findings.

≤ 15 findings — native multi-select (true checkboxes):

[ ] All P0 + P1
[ ] All fixable
[ ] [P0] [bug] Save button does nothing on Voice tab — F-a3f29b71
[ ] [P1] [a11y] Icon button missing label on Header — F-7c29f912
[ ] [P2] [ux] Toast disappears too fast on mobile — F-2bff1004
...

> 15 findings — bulk presets + custom indices:

1. All fixable
2. All P0 + P1
3. All P0 + P1 + P2
4. Custom selection (enter indices/IDs: 1,3,5-8 or F-a3f29b71)
5. Cancel

Either path confirms the resolved selection before proceeding.

Strategies

Batched PR (default for ≤ 3 findings)

All selected fixes applied sequentially in your current workspace. One PR at the end.

/candid-chrome-qa-fix --strategy batched

PR title (with issues): Fix N QA findings (ENG-1234, ENG-1235): <top finding title>. Body lists every finding with severity, surface, URL, and tracker link.

One PR per finding — parallel via Conductor

Emits a conductor:// deep link per selected finding and opens each one. Each link spawns a fresh Conductor workspace that fixes one finding and ships its own PR — fully in parallel.

/candid-chrome-qa-fix --strategy per-finding
/candid-chrome-qa-fix --strategy per-finding --print-links   # dry run

Conductor is Mac-only — on other platforms the skill falls back to --print-links automatically.

Local-only

Apply fixes to the working tree. No commit, no PR. Useful when you want to read the diff before deciding.

/candid-chrome-qa-fix --strategy local

Issues-only

Skip code changes entirely. File one tracker issue per selected finding. Useful for triage when you want to track findings without fixing them right away.

/candid-chrome-qa-fix --strategy issues-only

Issue Tracker Integration

When chromeQAFix.issueTracker.enabled is true (or --create-issues is passed), the skill files one issue per selected finding before any code change. Issue IDs are then embedded in commit messages, PR titles, and PR bodies so the tracker shows the PR as a linked attachment.

Linear (supported today)

{
  "chromeQAFix": {
    "issueTracker": {
      "enabled": true,
      "provider": "linear",
      "teamKey": "ENG",
      "state": "Backlog",
      "labels": ["qa-finding", "chrome-qa"],
      "priorityMap": { "P0": 1, "P1": 2, "P2": 3, "P3": 3, "P4": 4, "P5": 0 },
      "dedupe": true
    }
  }
}

Finding field	→ Linear field
`title`	Issue title (prefixed with `[QA]` )
`severity`	Priority (`priorityMap`; defaults: P0→Urgent, P1→High, P2/P3→Medium, P4→Low, P5→No-priority)
`category`	Appended to `labels`
`repro` / `expected` / `actual` / `suggestedFix`	Description (markdown)
`id`	Stored in description for deduplication on re-runs

Dedup probe

When dedupe: true, the skill calls mcp__claude_ai_Linear__list_issues for each finding and reuses an existing issue if its title or description contains the finding’s F-id. A second /candid-chrome-qa-fix run on the same findings file won’t create duplicates — it’ll reuse the prior issues.

Other providers

github, jira, and asana are stubbed for the future. They warn with a request-support link and skip the tracker step (code fixes still proceed). Set provider: "none" to disable tracker integration without removing the config.

CLI Flags

Flag	Description	Default
`--file <path>`	Use a specific findings file	(latest in `.context/findings/`)
`--severity P0,P1`	Only consider these severities	(all)
`--category bug,a11y`	Only consider these categories	(all)
`--strategy <name>`	One of `batched`, `per-finding`, `local`, `issues-only`	(prompted)
`--max-parallel N`	Per-finding deep-link launch batch size	`4`
`--print-links`	In per-finding mode, print deep links instead of opening them	`false`
`--fast`	In batched mode, hand off to `/candid-fast-ship` instead of `/candid-ship`	`false`
`--create-issues`	Force-enable tracker issue creation	(config)
`--no-create-issues`	Force-disable tracker issue creation	(config)
`--tracker-team <KEY>`	Override `issueTracker.teamKey` for this run	(config)

Configuration

Add a chromeQAFix block to .candid/config.json:

{
  "chromeQAFix": {
    "defaultStrategy": "batched",
    "maxParallel": 4,
    "testCommand": "npm test",
    "branchPrefix": "ron-myers",
    "conductorRepoPath": "candid-v1",
    "issueTracker": {
      "enabled": false,
      "provider": "linear",
      "teamKey": "ENG",
      "state": "Backlog",
      "labels": ["qa-finding", "chrome-qa"],
      "priorityMap": { "P0": 1, "P1": 2, "P2": 3, "P3": 3, "P4": 4, "P5": 0 },
      "dedupe": true
    }
  }
}

Field	Default	Description
`defaultStrategy`	`"batched"`	Used when no `--strategy` flag and selection > 3 findings
`maxParallel`	`4`	Per-finding deep-link launch batch size; sleeps 2s between batches
`testCommand`	`null`	Run after batched/local fixes; skipped if not set
`branchPrefix`	git user	Branch prefix for spawned-workspace branches (`<prefix>/qa-fix-<slug>`)
`conductorRepoPath`	derived	Value for the `path` query param of `conductor://` deep links
`issueTracker.enabled`	`false`	Master switch for the create-issue step
`issueTracker.provider`	`"linear"`	One of `linear`, `github`, `jira`, `asana`, `none`
`issueTracker.teamKey`	none	Required for Linear when enabled
`issueTracker.state`	`"Backlog"`	Initial state for new issues
`issueTracker.labels`	`["qa-finding", "chrome-qa"]`	Base labels; finding’s category is appended
`issueTracker.priorityMap`	see schema	Severity → Linear priority
`issueTracker.dedupe`	`true`	Search for existing F-id before creating

Hard Rules

The skill enforces these — it will refuse to act otherwise:

Never silently fix without user selection — always present the multi-select prompt unless explicit CLI flags target a known set.
Never open Conductor deep links on a non-macOS host — falls back to --print-links and warns.
Never write outside evidence.filesLikelyTouched without explicitly noting the extra files.
Never claim a fix is shipped without a PR URL.
Never assume v1 schema — validates schemaVersion === "2.0" and stops with a clear error otherwise.
Never URL-encode the deep link incorrectly — uses real URL-encoders, not hand-rolled string substitution.
Never spawn more than maxParallel Conductor workspaces in a single burst.
Never create a Linear issue without the single-issue invariant present in the rendered prompt.
Never silently fan out tracker creation — confirms before creating ≥ 5 new issues.
Never proceed with provider !== "linear" in v1 — warns and skips the tracker step. Code fixes still proceed.
Never embed secrets in issue descriptions, commit messages, or deep-link prompts.

Final Summary

After execution, the skill prints:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Candid Chrome QA Fix — Complete
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Findings file: .context/findings/2026-04-25-agent-config.json
Strategy:      batched
Selected:      4
Fixed:         3
Failed:        1
Skipped:       0
Issues filed:  4 created, 0 reused, 0 failed

Per finding:
  ✓ F-a3f29b71 [P0] Save button does nothing — PR: https://github.com/.../pull/123 | Issue: ENG-1234
  ✓ F-7c29f912 [P1] Icon button missing label — applied locally | Issue: reused ENG-1180
  ✗ F-2bff1004 [P1] Stale state on org switch — failed (test failure: line 42) | Issue: ENG-1235
  → F-1c00abcd [P1] Settings save loop — Conductor: ron-myers/qa-fix-settings-save-loop

Glyph legend: ✓ fixed/applied, ✗ failed, ◦ issues-only, → dispatched to Conductor.

Requirements

A candid-chrome-qa v2.0 findings file in .context/findings/ (run /candid-chrome-qa first if you don’t have one)
For Linear issue creation: the Linear MCP (mcp__claude_ai_Linear__save_issue) installed in your Claude Code environment
For per-finding parallel mode: macOS with Conductor installed (the skill falls back to --print-links on other platforms)
A writable .context/ directory (the skill creates .context/findings/<base>.fixes.md to log applied fixes)

FAQ

What if no findings file exists yet?

The skill prints No findings file found in .context/findings/ with a pointer to run /candid-chrome-qa first. No action is taken.

Can I re-run on the same findings file safely?

Yes. With dedupe: true (default), the Linear probe finds existing issues by F-id and reuses them rather than creating duplicates. The applied-fixes sidecar log (.context/findings/<base>.fixes.md) accumulates across runs.

What happens to findings with `severity: "P5"`?

P5 is reserved by /candid-chrome-qa for ✓ no issues — probes ran clean coverage receipts. The fix skill drops them automatically — they’re not bugs, they’re proof that probes ran.

Per-finding mode — what does each Conductor workspace do?

Each spawned workspace receives the full finding JSON, the desired branch name, and step-by-step instructions: rename branch → read files → apply fix → run tests → commit → push → ship via /candid-fast-ship. There’s no callback to the parent — check Conductor for progress.

Does the skill modify the findings file?

No. The findings file is the source of truth; the fix lifecycle is the consumer’s concern. The skill writes a separate sidecar log at .context/findings/<base>.fixes.md recording each fix applied.

Can I use this with a tracker other than Linear?

Not yet — only Linear is implemented in v1. github, jira, and asana are recognized config values that warn-and-skip with a link to request support. The provider abstraction is documented in the skill’s WORKFLOW.md so future implementations can plug in cleanly.

What’s the difference between this and `/candid-loop`?

/candid-loop runs /candid-review repeatedly until code-review issues are gone. /candid-chrome-qa-fix consumes a single Chrome QA findings file and routes each finding through fix + ship + (optional) tracker. They’re complementary: /candid-loop cleans the diff, /candid-chrome-qa-fix closes user-observable QA loops.