| name | opportunity-from-jd |
| description | Reads a raw JD and produces a YAML snippet for opportunities.yaml plus optional jd_catalog.csv row. Use when adding a new pipeline opportunity from data/raw, ingesting a JD file, or bootstrapping opportunity id and next_action fields. |
Add opportunity from a JD file
Role: Assistant — career / opportunity tracking (see docs/AGENT_ROLES.md).
Context to load:
- JD file: paste text or path to
data/<filename>.md
- Schema:
templates/opportunities_tracker_template.yaml
- Optional:
data/master.yaml for fit notes
- Optional:
config/jd_catalog.csv to register job_id
Scope:
- Needs only: JD, schema template, optional existing tracker/catalog rows.
- No broader vault context unless user asks.
Task:
- Read JD. Extract role_title and company_display (exact JD text). Company not in JD: don't guess — use user-provided label (e.g.
"Unknown (SF-based startup)"). No label either: ask first.
- Propose
id: short slug, unique in data/opportunities.yaml (e.g. acme-senior-backend-2026).
- Set
stage: usually Interested or Applied (user may override).
- Set
jd_source: min raw_path: "data/raw/<FILENAME>". If user wants catalog tracking, output one CSV row for jd_catalog.csv with columns job_id,alias,role,raw_path,normalized_path,status (fill unknowns with placeholder or ask).
- Set
next_action and next_action_date (concrete, within 7 days if possible), priority (P0/P1/P2), short notes.
- If user supplies contact info (recruiter, HM, referral), populate
contacts[] — each: name, role (recruiter|hiring_manager|referral|hr|other), channels (free map: any platform key → handle/URL/number, e.g. linkedin, email, zalo, whatsapp, telegram, messenger, phone), notes. Omit key if no contacts known.
- If JD or user provides portal/posting URL, populate
links (application_portal, job_posting, or free key). Canonical spots for manual status checks. Omit key if no URLs known.
- Set
updated_at to today (ISO date).
- Output YAML snippet ready to append under
active: in data/opportunities.yaml (or merge instructions if file exists). No duplicate id.
Output sections:
- YAML block for new
active[] item only
- If jd_catalog row suggested: CSV line or table row
- Assumptions / Risk
Constraints:
data/ is gitignored — full detail fine for YAML/chat drafts that stay local.- No fabrication. Only values from JD or user. No invented role details, company facts, contact info, URLs.
- Field can't be extracted: use
Unknown, note in Assumptions, ask one clarifying question before locking id or company_display.
- Every inference (stage from tone, priority from urgency language) → list in Assumptions, confirm before final output.
My inputs (collect, confirm, or ask):
- Any item below unknown from memory, prior messages, or repo (
data/raw/, opportunities.yaml): ask before emitting final YAML.
- Inferred item (stage from tone, path from last-opened file): state briefly, ask to confirm or correct before final output.
- Path to JD under
data/raw/: <...>
- Desired initial stage:
<Interested | Applied | ...>
- job_id for catalog (or "skip catalog"):
<...>
- Contacts (recruiter / HM / referral name + role + linkedin/email if known, or "none"):
<...>
- Application portal / job posting URL (or "none"):
<...>
