req-scan

name	req-scan
description	Scan publicly available sources for candidates who indicate they are open for hire. Uses WebFetch to read public APIs (HN Algolia, GitHub, dev.to). Writes prospect notes to Knowledge/Prospects/. Maintains cursor/dedup state in ~/.cache/fit/outpost/head-hunter/. Use when the head-hunter agent is woken or when the user asks to scan for open candidates.
license	Apache-2.0
metadata	{"version":"3.12.0","author":"forwardimpact"}

Scan Open Candidates

Fetch and filter publicly available candidate posts from platforms where people explicitly indicate they are open for hire. This skill handles fetching, filtering, deduplication, benchmarking, prospect-note writing, and memory updates.

Trigger

• The head-hunter agent is woken by the scheduler.
• The user asks to scan for open candidates or prospects.

Prerequisites

• WebFetch tool available (Claude Code built-in — no curl/wget).
• fit-pathway CLI available (bunx fit-pathway).
• Memory directory ~/.cache/fit/outpost/head-hunter/.

Inputs

• ~/.cache/fit/outpost/head-hunter/cursor.tsv — source rotation.
• ~/.cache/fit/outpost/head-hunter/seen.tsv — deduplication.
• Standard data via bunx fit-pathway skill --list and bunx fit-pathway job.

Outputs

• Knowledge/Prospects/{Name}.md.
• Updated cursor.tsv, seen.tsv, prospects.tsv, and log.md.
• ~/.cache/fit/outpost/state/head_hunter_triage.md.

<do_confirm_checklist goal="Verify the wake produced a clean, ethical scan">

• Selected the least-recently-checked, non-suspended source.
• Fetched via WebFetch (never curl/wget).
• Applied all 5 filters in order; passed candidates have ≥ 2 standard-relevant skills.
• Deduplicated against seen.tsv before processing.
• Benchmarked each prospect against a real bunx fit-pathway job.
• Prospect notes follow the template (no fabricated names).
• All state mutations went through node scripts/state.mjs — cursor, seen, prospects, failures, log.
• Triage report written to the state directory; alternative queries (if any) logged.
• Failures recorded; sources with ≥ 3 consecutive failures suspended.

</do_confirm_checklist>

Procedure

1. Pick a source

node scripts/state.mjs cursor list

Pick the least-recently-checked source. Skip suspended sources (failure get ≥ 3). If all are suspended, report it in the triage and exit. Source URLs and parse fields: references/sources.md.

2. Fetch and scan

Use WebFetch per references/sources.md. On failure, follow references/filters.md.

3. Filter

Apply the 5-filter pipeline in references/filters.md — signal, dedup, geographic, skill alignment, experience level.

3b. Fallback — zero new prospects

If filtering eliminates every candidate, try up to 3 alternatives per wake from references/fallbacks.md. Log every alternative in log.md.

4. Benchmark and write prospect notes

For each candidate that passes filters, benchmark:

bunx fit-pathway job {discipline} {estimated_level} --track={best_track}

Classify match strength:

• strong — multiple core skills match, level aligns, location works, plus non-traditional signals for forward-deployed.
• moderate — some overlap, level roughly right, minor gaps.
• weak — few matching signals, significant gaps.

Write notes for strong and moderate matches only, using the template in references/template.md.

mkdir -p "Knowledge/Prospects"

5. Update state

All state changes go through node scripts/state.mjs — full command reference: references/state.md. Each wake:

1. Update the cursor (cursor set).
2. Reset failure count on success or increment on failure.
3. Append every processed post ID to seen (seen batch for many).
4. Add new prospects (prospect add).
5. Append the wake summary (log-wake).

6. Write the triage report

Save to ~/.cache/fit/outpost/state/head_hunter_triage.md:

# Head Hunter Triage — {YYYY-MM-DD HH:MM}

## Last Scan
Source: {source_id} ({description})
Posts scanned: {N}
New prospects: {N}
Skipped: {N} (dedup: {N}, location: {N}, skill fit: {N})
Alternative queries tried: {N} ({list, or "none needed"})

## Pipeline Summary
Total prospects: {N} (strong: {N}, moderate: {N})
Sources checked today: {list}
Oldest unchecked source: {source_id} (last: {date})
Suspended sources: {list with failure counts, or "none"}

## Recent Prospects
- **{Name}** — {match_strength}, {estimated_level} {track}, {location}

## Retention
{List prospects older than 90 days not acted on, if any.}