career-role-enrich

name	career-role-enrich
description	Enrich a career-changelog role block and sync the matching resume.json work entry. Walks intake → audit → draft → review → apply for one role at a time.
argument-hint	<role-slug>
allowed-tools	Read, Edit, Write, Bash(grep *), Bash(git status), Bash(git branch --show-current), AskUserQuestion

Career Role Enrich

Project-local skill for enriching one role at a time in _includes/career/changelog.html and the matching resume.json work entries. Captures the SSENSE-iteration workflow as a reusable process so future role rounds (Deliverr, Flexport, Octav, Shopify Logistics, Shopify) can ride the same rails without re-deriving conventions.

Input: $ARGUMENTS (a role slug like deliverr, flexport, octav, shopify-logistics, shopify)

Step 0: Validate Input

If $ARGUMENTS is empty, STOP and ask the user for a role slug. Recommended slugs match the role's display name with lowercase + hyphenated transformation (e.g., Shopify Logistics → shopify-logistics).

The slug MUST match ^[a-z0-9-]+$. If it contains anything else (spaces, slashes, shell metacharacters), STOP and ask for a clean slug.

Step 1: Resolve Target

Use the slug to locate three things:

1. Changelog block in _includes/career/changelog.html. The block is the <h2 id="changelog-<version>"> heading and everything until the next <h2> (or the closing </div>). To find it, grep for the role's display name (e.g., Deliverr) and walk up to the nearest <h2>.
2. Resume entries in resume.json. Grep for "name": "<role-display>". There may be multiple entries per role (SSENSE has Tech Lead + Senior Developer; Vidéotron has two; AFS has two). All entries with the matching name are in scope.
3. Published PDF. The source of truth is https://ouimet.info/charles/resume/Charles%20Ouimet%20-%20Principal%20Developer%20-%20Resume%202026-05.pdf. Read it via the Read tool when needed for cross-referencing.

Record the changelog block's line range and the resume entries' line ranges for later steps.

Step 2: Audit

Three-way compare: PDF role sub-section × changelog block × resume entries.

For each PDF bullet, classify as:

• Already covered: present in both changelog and resume in equivalent form.
• Missing from changelog: present in resume only.
• Missing from resume: present in changelog only.
• Stale wording: present but using outdated phrasing (e.g., Use of more instead of Use of additional, Technical Architecture Group instead of Technology).

Produce a /note describing the audit table. Use description audit-<role-slug>. The audit /note is the input to Step 3 intake decisions.

Step 3: Intake

Use /question to gather context the audit can't infer from existing files. The question file uses description intake-<role-slug> and includes these reusable prompts as Q001-Q005:

1. Headline projects or initiatives at this role. For each: a one-line "what" and a one-line "what was hard about it."
2. Scope and scale. Team size, services owned, request volumes, deploy cadence, geographic scope, departments coordinated with.
3. Shout-outs. Collaborators, mentors, key allies. LinkedIn URLs help.
4. Problems solved framed as outcomes, not tools. Pick 2-4 problems where the outcome matters and the tool was the means.
5. Anything missing from the PDF / changelog / resume. Internal talks, mentorship, hiring contributions, incidents handled, etc.

Wait for the user to fill in answers before proceeding to Step 4.

Step 4: Draft

Produce a plan /note (description plan-<role-slug>) with concrete proposals for:

1. Changelog block HTML: full proposed text for the role's Added block (and Changed / Deprecated if applicable). Apply career-style conventions throughout (see "Conventions consulted" below).
2. Resume entries: proposed summary text and highlights[] array for each resume.json:work[] entry under this role.
3. Skills additions (if any): tools introduced at this role that need to be added to resume.json:skills. Do NOT re-add tools already in skills from earlier roles.
4. Decision points flagged for Step 5 review: anything ambiguous goes into one or more /question files.

Consult /prose-style (hard-wrap, code refs, GitHub refs) and /humanizer (em-dash sweep, AI-vocabulary check) before writing the plan note.

Step 5: Review Iteration

For each decision cluster, create a /question file (description <cluster-name>-<role-slug>). Common clusters from the SSENSE precedent:

• Bullet structure: nested <ul> vs flat; sub-bullet split points; lead-in content.
• Tool fold-in: fold standalone Use of X bullets into a project bullet (matches OMS's state machine / Step Functions / event storming fold-in pattern) vs keep standalone.
• ID assignment: which new project bullets get changelog-<role>-added-<slug> IDs.
• Trim vs keep: existing content that overlaps with new additions.
• Summary scope for resume.json: which summary fields get refreshed; what they say.

The user edits answers in the question files; the skill reads them back and iterates. Repeat until all decisions resolve.

Step 6: Apply

Edit _includes/career/changelog.html and resume.json per the agreed plan.

After applying, run post-edit verification:

1. JSON validity: python3 -c "import json; json.load(open('resume.json'))". Must succeed.
2. Additive convention sweep: grep -n "Use of more" _includes/career/changelog.html. Should return empty. Any match is a stale-wording bug; rename to Use of additional.
3. Em-dash sweep within the role's block: extract the block's line range from Step 1, then sed -n "<start>,<end>p" _includes/career/changelog.html | grep -n "—". Should return empty.
4. No <strong> micro-headings inside sub-bullets: grep -n "<li><strong>" _includes/career/changelog.html. Should return empty.
5. First-person verb sweep within the role's block (excluding Key relationships at <Role>: sub-bullets, which are allowed first-person voice per career-style Rule 3 exception): sed -n "<start>,<end>p" _includes/career/changelog.html | sed '/<li>Key relationships at .*:/,/<\/ul>/d' | grep -nE "\bI (built|led|created|maintained|developed|designed|implemented|authored|introduced|oversaw|delivered|shipped|wrote)\b". Should return empty. Matches indicate narrative voice that needs reshaping into noun-phrase form. The middle sed deletes the Key-relationships parent through its closing </ul> so the first-person exception scoped to those sub-bullets doesn't produce false positives.
6. Vague-filler sweep within the role's block: sed -n "<start>,<end>p" _includes/career/changelog.html | grep -nE "(and more[.,]|\betc\.\s)". Investigate matches; and more is almost always cuttable; etc. outside parenthetical examples is usually filler.
7. Rigorous K-a-C re-read (manual): re-read the edited block as a stranger encountering this changelog cold. For each <li>, ask: is this a fact, or is it narrative/editorial framing? Flag any sentence whose value is context or justification rather than what was added/changed/done. Iterate with the user if anything looks off.

Step 7: Wrap Up

Print a status report:

=== Role '<slug>' enrichment applied ===

Audit:        <path-to-audit-note>
Intake:       <path-to-intake-question-file>
Plan:         <path-to-plan-note>
Decisions:    <paths-to-decision-cluster-question-files>
Files edited: _includes/career/changelog.html, resume.json

Next: invoke /finish-issue when the branch is ready to ship.

The skill does NOT invoke /finish-issue automatically. The user runs it when ready.

Conventions consulted

• /prose-style — hard-wrap rule, code reference syntax, GitHub reference syntax. Auto-consulted when writing any file content.
• /humanizer — em-dash overuse, AI vocabulary words, passive voice, filler phrases.
• career-style (project-local, non-invocable) — additive Use of X convention, role-scoped IDs, role-voice. Auto-consulted by this skill.
• career-layering (project-local, non-invocable) — three-layer information architecture (changelog → resume.json → docx/PDF), distillation rules, cross-file consistency. Auto-consulted by this skill.

Quality Checklist

Before reporting Step 7:

• Resolved target (changelog block + resume entries + PDF reference)
• Audit /note produced
• Intake /question answered
• Plan /note produced and reviewed
• Decision-cluster /question files resolved
• Changelog edits applied; additive-convention sweep returns empty
• Resume edits applied; JSON validates
• Em-dash sweep within role block returns empty
• No <strong> micro-headings in sub-bullets
• First-person verb sweep returns empty (excluding Key relationships at <Role>: sub-bullets)
• Vague-filler sweep investigated (no remaining and more / loose etc.)
• Rigorous K-a-C re-read complete (no narrative/editorial sentences remaining)
• Skills additions reviewed (no duplicates from earlier roles)