跟踪器

---
name: tracker
description: Provider-neutral contract for every issue-tracker and code-host action (create/update/comment/close issues, attach files, comment on MRs/PRs, read CI runs and logs) plus label conventions and failure handling. Every WAY calls these operations; the matching provider-* skill says how to perform them on GitLab, GitHub, Jira, Azure DevOps, or Bitbucket.
---

# tracker

WAYs never talk to a provider directly. They call the **operations** below,
and the matching `provider-<name>` skill says how to perform each one. Skills
and agents that write content (e.g. `bug-report` formats a bug and picks its
labels) own the content; this skill owns the mechanics.

## 1. Pick the providers (every WAY, once per session)

Two settings in the project's `.env` (set by `qa-setup`):

| Setting | Values | Covers |
|---|---|---|
| `QA_TRACKER` | `gitlab` · `github` · `jira` · `azure-devops` | issues: bugs, test plans, test cases, stories, reports |
| `QA_CODE_HOST` | `gitlab` · `github` · `azure-devops` · `bitbucket` | MRs/PRs, CI runs and logs, CI config |

Read them **without exposing secrets**. Never `cat` or `Read` the whole `.env`, because it can hold tokens. Use:

```bash
grep -E '^(QA_TRACKER|QA_CODE_HOST|GITLAB_PROJECT_ID|GITHUB_REPO|JIRA_BASE_URL|JIRA_PROJECT_KEY|AZDO_ORG_URL|AZDO_PROJECT|AZDO_REPO|BITBUCKET_WORKSPACE|BITBUCKET_REPO|STAGING_URL|PROD_URL|FEATURE_BASE_URL|API_PERF_THRESHOLD_MS)=' .env
```

If they're unset, fall back like this: a connected GitLab MCP server means `gitlab`, a `github.com` git remote means `github`. Otherwise ask, and suggest `qa-setup`. Then load `provider-<QA_TRACKER>` (and `provider-<QA_CODE_HOST>` if it's different) and follow it for every operation below.

## 2. Vocabulary

WAY text uses neutral words. Each provider maps them:

| Neutral term | GitLab | GitHub | Jira | Azure DevOps | Bitbucket |
|---|---|---|---|---|---|
| issue | issue | issue | issue (Bug / Story / Task) | work item (Bug / User Story / Task) | — use Jira |
| issue ref | `#143` | `#143` | `PROJ-143` | `#143` (work item ID) | — |
| change request | merge request `!87` | pull request `#87` | — | pull request `!87` | pull request `#87` |
| CI run | pipeline | workflow run | — | pipeline run (build) | pipeline |
| milestone | milestone | milestone | fixVersion / sprint | iteration path | — |
| label | label | label | label | tag | — |

Always report refs back in the provider's own format (`#143`, `PROJ-143`, `!87`).

## 3. Operations

| Operation | Used by WAY | Notes |
|---|---|---|
| `issue.create(title, body, labels)` → ref + URL | 1, 2, 3, 4, 7, 8, 12, 13 | body is Markdown; the provider skill converts it if the tracker needs another format |
| `issue.get(ref)` → title, body, labels, state, milestone, links | 1, 4, 6, 9, 14 | |
| `issue.update(ref, body?, labels+?, labels-?)` | 6, 9 | body is a **full replace**: write the watermark again every time |
| `issue.comment(ref, body)` | 1, 6, 9 | |
| `issue.close(ref, comment)` / `issue.reopen(ref, comment)` | 9, 13 | outward-facing: only when the WAY says so, always with a comment |
| `issue.list(filters)` | 9, 12, 14 | filters: labels, milestone, state, text search |
| `issue.link(ref, other_ref)` | 1, 4 | test-plan ↔ test-case, bug ↔ test-case |
| `file.attach(path)` → Markdown snippet or link | 1, 2, 3, 9 | see §5 |
| `change.get(ref)` → title, description, source branch, linked issues, changed files | 5, 11 | |
| `change.diff(ref, files?)` | 5, 11 | only fetch the files you need |
| `change.comment(ref, body)` | 1, 4, 5, 11 | |
| `ci.get_run(ref or URL)` → status, branch, commit, trigger, jobs | 5 | |
| `ci.job_log(job, tail=200)` | 5 | read the **end** first, page back only if needed |
| `ci.latest_run(branch)` | 12 | |
| `ci.lint(config)` | 10 | not every provider has one: the provider skill says what to do instead |
| `project.check()` | 0 | proves the credentials reach the project |

If the provider skill marks an operation as unsupported, say so in the WAY's output and carry on. Never fake it.

## 4. Label conventions

Draw labels only from this set, unless a WAY says otherwise. On Jira they're labels (severity/priority also go in the Priority field), on Azure DevOps they're tags (plus the Severity/Priority fields). See the provider skill.

| Label | Meaning | Set by |
|---|---|---|
| `bug` | Any filed defect | WAY 1/2/3 |
| `sev-1` / `sev-2` / `sev-3` / `sev-4` | Severity, from the `bug-report` skill | WAY 1/2/3 |
| `p1` / `p2` / `p3` | Priority, from the `bug-report` skill | WAY 1/2/3 |
| `test-plan` | Index issue summarizing a set of test cases | WAY 1/4 |
| `test-case` | One issue per generated `Test#<nn>` | WAY 1 |
| `api` | Bug or test case from API testing | WAY 2 |
| `e2e` | Bug or test case from browser testing | WAY 1 |
| `accessibility` | Bug from an axe-core violation | WAY 1 |
| `qa-session-report` | End-of-session summary issue | WAY 1/2 |
| `benchmark` | Competitor benchmark report, or a gap backlog index | WAY 7/8 |
| `user-story` | User story written from a benchmark gap | WAY 8 |
| `benchmark-gap` | Story closing a feature gap vs a competitor | WAY 8 |
| `decision-needed` | Story that conflicts with a deliberate product choice | WAY 8 |
| `verified` | Bug whose regression test passed after the fix | WAY 9 |
| `fix-failed` | Bug whose regression test still fails after a claimed fix | WAY 9 |
| `performance` | Bug from a load-test threshold breach | WAY 15 |
| `flaky-test` | Test (not product) issue: a quarantined flaky scenario | WAY 13 |
| `release-readiness` | Go/no-go report for a milestone | WAY 12 |

If a label doesn't exist yet, the provider skill says whether to create it or just apply it.

## 5. Attachments

A local path like `runs/<id>/artifacts/…/test-failed-1.png` means nothing to a developer. Attach the file (`file.attach`) and put the returned snippet in the body. Attach only what helps reproduce: the failure screenshot, plus `error-context.md` or a short console excerpt. Never attach traces (`trace.zip`) or `api-responses/` captures unless you've checked they hold no tokens or personal data. If attaching fails or isn't supported, keep the local path, label it "on the QA machine", and carry on.

## 6. Failure handling

If any provider call fails (auth, network, rate limit, not found):
1. Log what was attempted and what the provider returned
2. Skip that operation. Don't retry in a loop
3. Continue the rest of the WAY, and don't abort

If that call was the only way to deliver something a human needs (e.g. an enhanced story body), print the content to chat so the user can apply it by hand.

## Rules
- Never invent a project ID, repo, key or ref. If a required setting is missing, stop and ask (or suggest `qa-setup`)
- **Secrets stay out of the conversation.** Pass tokens only as environment variables inside the command that needs them (`set -a; . ./.env; set +a; curl -u "$JIRA_EMAIL:$JIRA_API_TOKEN" …`). Never echo them, and never put them in a body, URL or file
- Every human-facing body and comment carries the watermark footer from CLAUDE.md's Output Watermark. The caller includes it. This skill doesn't add it
- Write bodies to a temp file and pass it by path (`--body-file`, `-d @file`), not as a shell argument. That avoids quoting and length problems