Vvanlaar/ado-investigate

Investigate ADO Ticket

Fetch ADO ticket → score bluebillywig GitHub repos → update Custom.Repository field → append AI analysis to description.

Read ~/.claude/skills/_ado-shared.md for auth, env vars, fetch ticket, write-payload, and Windows/MSYS2 patterns.

Workflow

1. Fetch ticket + current field values
2. Fetch non-archived repos from bluebillywig org
3. Score repos against ticket content
4. Read READMEs for uncertain top candidates
5. Update ticket: Repository field + description appendix

---

Step 1: Fetch Ticket

Use fetch with $expand=all (see shared). Extract and note:

• System.Title
• System.Description (HTML — strip tags for keyword analysis)
• System.WorkItemType, System.AreaPath
• Custom.Repository — preserve existing values when updating
• Microsoft.VSTS.Common.AcceptanceCriteria if set

---

Step 2: Fetch GitHub Repos

Skip this step — docs/bluebillywig-repos.md already has all active repos. Only fetch live if you suspect the list is stale (>2 weeks old):

gh repo list bluebillywig --limit 100 --json name,description,repositoryTopics,isArchived \
  > /tmp/repos-raw.json
node << 'NS'
const raw = require('fs').readFileSync(require('os').tmpdir() + '/repos-raw.json', 'utf8');
const filtered = JSON.parse(raw).filter(r => !r.isArchived);
console.log(JSON.stringify(filtered, null, 2));
NS

---

Step 3: Score Repos

First: read C:\dev\ai\orch\docs\bluebillywig-repos.md — describes every active repo with keyword→repo scoring guide.

Extract keywords from ticket (title + stripped description + area path):

| Signal | Points | |--------|--------| | Repo name word in title | +40 | | Repo name word in description | +20 | | Topic matches ticket keyword | +20 each | | Description word overlap | +5 per keyword | | Area path segment matches repo name | +15 |

Include repos scoring >30 pts, max 3. If top candidates <60 pts, fetch README:

gh repo view bluebillywig/<repo> --json readme --jq '.readme.text' | head -c 3000

---

Step 4: Update Repository Field

Custom.Repository is semicolon-separated. Merge with existing — never overwrite blindly.

node << 'NS'
const fp = require('path').join(require('os').tmpdir(), 'ado-payload.json');
const existing = 'EXISTING_VALUE_OR_EMPTY';   // from Step 1
const toAdd = ['repo-a', 'repo-b'];
const merged = [...new Set([...existing.split(';').filter(Boolean), ...toAdd])].join(';');
require('fs').writeFileSync(fp, JSON.stringify([
  { op: 'replace', path: '/fields/Custom.Repository', value: merged }
]));
NS

Then PATCH via shared write-payload pattern.

---

Step 5: Append Analysis to Description

Append to System.Description — never replace the full existing content.

<hr>
<p><strong>[AI Investigation – YYYY-MM-DD]</strong></p>
<p><strong>Relevant repos:</strong></p>
<ul>
  <li><strong>repo-name</strong> — reason (e.g. "name matches 'player', has 'video' topic")</li>
</ul>
<p><strong>Suggested approach:</strong> 1–2 sentence implementation hint.</p>

Then PATCH via shared write-payload pattern (append to existing description HTML).

---

Output to User

## Investigation: ADO #<ID> — <title>

**Repos identified:**
- repo-a (85% — "player" in name + video topics)
- repo-b (60% — area path "Core" matches)

**Ticket updated:**
- ✅ Custom.Repository: repo-a;repo-b
- ✅ Description: AI analysis section appended

**Suggested approach:** <1–2 sentences>

---

Common Mistakes

| Mistake | Fix | |---------|-----| | Overwriting existing repos | Read Custom.Repository first, merge with new values | | Replacing full description | Fetch current System.Description, append only | | Including archived repos | Filter isArchived == false from gh repo list | | Guessing repo names | Names from gh repo list are the exact picklist values — use as-is | | topics field in gh repo list | Wrong field name — use repositoryTopics | | --order flag in gh repo list | Flag doesn't exist — omit it |