santifer--career-ops/modes/add.md

Mode: add — Add a project, paper, or role to your CV

Fetch a finished project / paper / internship from a link (or plain text), turn it into ATS-style CV content grounded only in what the source actually says, preview it, and — after you confirm — append it to cv.md and (for projects) article-digest.md. Deterministic dedup and insertion are handled by add-entry.mjs, so re-adding the same thing is a safe no-op.

Non-negotiables (from the project's source-of-truth rules in _shared.md):

• Confirm before write. Never touch cv.md / article-digest.md until the user approves the preview.
• Never fabricate. Every bullet, metric, and date must be backed by the fetched page or the user's own words. If the source doesn't state it, it does not go in. Keywords get reformulated, never invented.
• Zero-key, local. Use only public, no-auth fetches (GitHub's public API, WebFetch). No API keys, no third-party services.

Input

$mode after add is the source. Accept any of:

• a GitHub repo URL (github.com/<owner>/<repo>)
• a paper / publication link (arXiv, DOI, journal, personal page)
• a project / portfolio page URL
• plain text the user pastes describing the work

If no source was given, ask the user for one.

Pipeline

1. Load context. Read cv.md (its existing section names and formatting are the template to match) and article-digest.md if present.

2. Fetch the source (zero-key):

   • GitHub repo → the public REST API (https://api.github.com/repos/<owner>/<repo> for name/description/topics/language/stars/timestamps) plus the README via WebFetch. No token required for public repos.
   • Any other link → WebFetch. Only fall back to Playwright if the page is JS-rendered and WebFetch returns nothing useful.
   • Plain text → use it directly as the source; do not invent beyond it.

3. Extract structured facts actually present in the source: name, dates / period, tech stack, role, and concrete outcomes/metrics. Leave anything the source doesn't state blank — do not guess.

4. Classify the entry type → target CV section (see table). Placement is inferred, but shown in the preview; only ask the user when it's genuinely ambiguous.

5. Write ATS bullets from the extracted facts — 2–4 concise, quantified-where-the- source-supports-it bullets, matching the bullet style already used in cv.md. For a project, also compose an article-digest.md block (## <Name> — <tagline> with **Hero metrics:**, **Architecture:**, **Key decisions:**, **Proof points:**), filling only what the source supports.

6. Preview. Show, as a diff-style preview: the inferred CV section, the exact markdown to be inserted into cv.md, and (for projects) the article-digest.md block. Flag anything you could not source.

7. Confirm gate. Ask the user to approve, edit, or cancel. Do not proceed without an explicit yes.

8. Write via the helper. Build the payload (schema below), write it to /tmp/add-<slug>.json, then run:

   node add-entry.mjs /tmp/add-<slug>.json
   

   (Add --dry-run first if the user wants to see the file-level change without writing.)

9. Report the helper's JSON result. If a target comes back duplicate, tell the user it was already present and nothing was changed.

Section inference

Source is…	CV section
a code project / repo / tool	Projects
a paper / publication / preprint	Publications (create if absent)
an internship / job / role	Work Experience
a talk / course / certification	Education (or ask if unclear)

add-entry.mjs creates the section heading if it doesn't exist yet, so a new ## Publications is fine.

Payload schema (input to add-entry.mjs)

Both keys optional; provide at least one. articleDigest is for projects only.

{
  "cv": {
    "section": "Projects",
    "dedupKey": "<short canonical name, e.g. the repo/project name>",
    "entry": "<exact markdown to insert — a bullet for Projects, or a\n### Org — Location / **Title** / dates / bullets block for Work Experience>"
  },
  "articleDigest": {
    "dedupKey": "<same canonical name>",
    "entry": "## <Name> — <tagline>\n\n**Hero metrics:** ...\n\n**Architecture:** ...\n\n**Key decisions:**\n- ...\n\n**Proof points:**\n- ..."
  }
}

dedupKey is normalized (case- and punctuation-insensitive) to detect an entry that's already there, so the command is idempotent.

Rules

• Match the existing cv.md formatting exactly (heading levels, bullet style, date format) — the file is the template.
• One entry per run. To add several, run add per item.
• If the fetch fails or the page has no usable content, say so and stop — never synthesize an entry from nothing.
• Personal data (the CV itself) stays local; the fetch only reads public sources.