Ingest Resume History

This reads everything in inbox/ and writes plain-text versions to inbox/.extracted/.

2. Read everything

• All files in inbox/.extracted/
• Current content/master.yaml (if it exists; may be the example)

3. Propose a reconciled master

Build a proposed master.yaml that:

• Collects roles across all sources, keyed by (company, title, start) for dedupe.
• Merges bullet variants for the same role: prefer longer/more specific wording; flag conflicts to the user.
• Assigns stable IDs. ID scheme: <company-slug>-<startYear> for roles; <role-id>-<keyword> for bullets.
• Preserves any existing IDs in the current master.yaml exactly — do not renumber.

Present the proposal as:

• A list of new roles being added (with reasoning: "from Acme-2023.pdf").
• A list of new bullets per existing role.
• A list of possible duplicates you'd like confirmation on.

4. Apply on approval

On explicit user approval, write the full proposed master.yaml. Commit:

git add content/master.yaml
git commit -m "Ingest history from inbox"

5. Archive originals

Move processed PDFs from inbox/ to jobs/archive/<inferred-company>/ (create folder if needed). For each moved PDF, create a stub job.md next to it with whatever context you can infer from filename or content. If inference is weak, write "inferred context TBD" and flag to the user.

Editing discipline

• When re-running ingest on new files, update master incrementally — do not regenerate from scratch.
• If a source file contradicts existing master content (e.g., different dates for the same role), raise the conflict to the user rather than picking a winner.