Rick Intel

Rick Intel

burp Before you write a line of code, you read the dossier, Morty. This skill reads the ADO work item, climbs to the parent epic, sweeps vexis-docs and ADRs, and writes a story-scoped intel file at docs/rick/<folder>/intel/current.md.

Sister of rick-plan. Plan tells you what to build. Intel tells you what's already known. Run intel first.

Argument

Required. The ADO work item id. Accepts AB16289, 16289, or ab16289. Normalize to AB<digits>.

If $ARGUMENTS is empty or no AB-id can be parsed, print one line and stop:

Fine, Morty, which story. Pass an ADO id.

Optional --refresh anywhere in $ARGUMENTS triggers re-gather (see Refresh).

Intro

Pick one line at random from INTROS.md, substitute <AB-id>, print it. Then proceed.

Pre-flight

1. Normalize the AB-id.
2. Resolve the folder. Intel is almost always run pre-branch (the whole point is research before you build), so the folder name should match the branch the user will create after reading. Priority order, first that produces a value:
   1. Branch-already-cut override. git branch --show-current returns a feature branch (not in {main, master, develop} and non-empty) AND the branch name contains the AB-id digits (e.g. branch feat/15920-data-gap for AB15920) → sanitize (/ and spaces → -) and use it. The digit-substring check is mandatory — without it, running /rick-intel AB15920 while sitting on feature/WP-FE7-Template-Management lands the dossier inside the unrelated feature's folder. If you're on a feature branch for a different story, priority 1 doesn't fire.
   2. Predict the branch name. Fetch ADO title cheaply:

      az boards work-item show --id <digits> --org https://dev.azure.com/VizyFintech --query 'fields."System.Title"' -o tsv
      

      Slugify per rick-plan/PICK-FOLDER.md step 2: truncate at the first em-dash / en-dash / -; lowercase; replace non-[a-z0-9] runs with -; trim trailing hyphens; cap at 40 chars without cutting mid-segment. Result: <digits>-<title-slug> (e.g. 16289-rotate-share-link-ui).
   3. ADO title fetch failed → fall back to ab<digits> (e.g. ab16289) and record an open question.
3. Compute paths:
   • Folder: docs/rick/<folder>/intel/
   • Canonical: docs/rick/<folder>/intel/current.md
4. Glob check for prior intel at this AB-id (handles the case where intel was written before but the folder name has since drifted, e.g. a different branch was created later): find docs/rick -type f -path '*/intel/current.md' -exec grep -l '<AB-id>' {} \;. If a single match exists at a different path, treat it as the canonical instead (don't write a second copy under the predicted name). If multiple match, print them and stop — Morty resolves.
5. Check if canonical exists:
   • Exists, no --refresh → print path + the file's Last gathered: line + the one-line summary footer, stop. Do not re-fetch.
   • Exists, --refresh → load existing findings into memory for diff (see Refresh).
   • Does not exist → fresh gather. Cache the title from step 2 so the gather phase doesn't re-fetch it.
6. Create docs/rick/<folder>/intel/ if missing.

Gather (no code, info only)

Source order: ADO → vexis-docs → ADRs. Record any failure as an open question and keep going.

1. ADO work item

az boards work-item show --id <digits> --org https://dev.azure.com/VizyFintech

Extract:

• fields.System.Title → story title
• fields.System.Description → raw description (HTML — strip tags for Scope prose, keep plaintext for AC parsing)
• fields.Microsoft.VSTS.Common.AcceptanceCriteria → AC field if present
• fields.System.State, fields.System.IterationPath, fields.System.AssignedTo.displayName
• relations[?attributes.name=='Parent'] → parent work item URL → parent id
• relations[?attributes.name=='GitHub Pull Request'] → PR refs (extract last URL segment as PR number)

If az fails or unauthenticated, record open question "ADO unauthenticated — run az login and --refresh" and continue with whatever else is reachable.

2. ADO parent epic

For the parent id from step 1, run az boards work-item show again. Extract title and AC. Use to derive the domain anchor and to quote parent AC text when the story Backs AC-N of parent.

3. AC parsing

Try each source in order, stop at first that yields ACs:

1. fields.Microsoft.VSTS.Common.AcceptanceCriteria field on the story.
2. Story description body — scan the HTML-stripped description for AC content:
   • "Acceptance Criteria" heading (any level, case-insensitive; also "AC", "Criteria") followed by a list.
   • AC-?\d+[:.\)] markers with the prose that follows.
   • Gherkin-style Given … When … Then … blocks.
3. Parent epic's AC field — when story description references AC-N (regex AC-?\d+ or Backs AC: (AC-?\d+(?:, AC-?\d+)*)), pull the referenced text from parent.

No AC parseable → file uses one ## Findings flat section instead of per-AC sections. Everything else is unchanged.

4. Domain anchor

Derive candidate tokens from parent title + story title:

• Lowercase, tokenize on spaces, |, -, /.
• Drop noise: BE, FE, WP, digits, service, module, task, bug, epic, story, generic verbs.

Find docs root, first that exists wins:

1. vexis-docs/ in current repo
2. ../vexis-docs/ sibling repo
3. docs/ fallback

For each token: find <docs-root> -type d -iname '*<token>*'. First directory match wins. Full read of that subtree.

No match → record open question "No domain anchor matched for tokens: …". Continue with sweep only.

5. Keyword sweep

Extract sweep terms from story title + Scope text + AC titles. Drop noise tokens.

For each term: grep -rli --include='*.md' <term> <docs-root> docs/adr/ 2>/dev/null.

Dedupe hits against the domain-anchor read. For each unique hit: open the file, find the first paragraph containing the term, summarize in ≤2 sentences. Finding text = that summary. Source = <path>:<line> of the matched paragraph.

Terms with zero hits → open question "No vexis-docs match for <term>".

6. ADRs

docs/adr/ (or vexis-docs/adr/ if exists). Same sweep over titles and bodies. Same finding format.

Write the file

Path: docs/rick/<folder>/intel/current.md. See OUTPUT.md for the full section order, finding format, and refresh stale-line format.

Refresh

When --refresh is passed and the file exists:

1. Read current.md, parse every - … — Source: <S> — Found: <D> line into a map keyed by source.
2. Run the full gather phase fresh.
3. For each gathered finding:
   • Same source, summary matches → keep existing line untouched.
   • Same source, summary differs → strike old, append new with today.
   • New source → append new with today.
4. For each existing source not present in fresh gather → strike old with today as Stale.
5. Update Last gathered: to today.
6. Rewrite the file.

Never delete a finding line. Stale markers are history.

ADO writeback

After writing the file, post an HTML comment to the ADO work item. Body shape + idempotency rule live in ADO-COMMENT.md. Substitute <AB-id>, <N>, and <YYYY-MM-DD> before posting.

Failure → don't block file write. Print:

Warning: ADO writeback failed: <reason>
Run manually: az boards work-item update --id <digits> --discussion '<html-body>'

bd writeback

After writing the file, attempt bd linkage.

1. bd list --json 2>/dev/null — bd absent → skip silently.
2. Find issue where metadata.ado == <id-numeric>.
3. Exactly one match → bd update <bd-id> --set-metadata research=docs/rick/<folder>/intel/current.md.
4. Zero matches → print: No bd issue linked. Once claimed: bd update <bd-id> --set-metadata research=docs/rick/<folder>/intel/current.md. Continue.
5. Multiple matches → print candidate ids, do nothing, ask Morty to resolve.

Confirm and stop

Fresh write:

Intel filed for <AB-id>: docs/rick/<folder>/intel/current.md
<N> ACs, <N> findings, <N> open questions
Domain anchor: <path> (<N> hits) | Keyword sweep: <N> cross-cutting hits

If the folder was predicted (Pre-flight step 2 priority 2 — no feature branch was checked out) AND the predicted name differs from the current branch, follow up with one line so Morty can branch into the dossier:

Next: git checkout -b <folder>

Refresh:

Intel refreshed for <AB-id>: docs/rick/<folder>/intel/current.md
<N> carried, <M> updated/stale-marked, <K> new

If ADO writeback warned, print the warning line first.

Rules

• No code. No implementation suggestions. Pure research synthesis.
• Cite every finding. Path-line or URL. No bare claims.
• Verbatim ADO description in Scope — strip HTML, don't paraphrase.
• Stale-mark, never delete.
• Idempotent. Two runs without --refresh should be free: read, print, exit. Don't re-fetch ADO unprompted.
• burp Don't suit up twice for the same dimension, Morty.