talent-sourcing（人才采购）

Talent Sourcing

Candidate discovery powered by Nimble Web Search Agents.

User request: $ARGUMENTS

Before running any commands, read references/nimble-playbook.md for Claude Code constraints (no shell state, no &/wait, sub-agent permissions, communication style).

Instructions

Step 0: Preflight

Run the preflight pattern from references/nimble-playbook.md (5 simultaneous Bash calls: date calc, today, CLI check, profile load, index.md load).

Also simultaneously:

• mkdir -p ~/.nimble/memory/{reports,talent-sourcing}

From the results:

• CLI missing or API key unset → read references/profile-and-onboarding.md, stop
• Profile exists → note industry keywords if any; proceed to Step 1
• No profile → fine, talent-sourcing doesn't require onboarding; proceed to Step 1

Step 1: Parse Request & Confirm Search Parameters

Parse $ARGUMENTS for:

• Role — job title or function (e.g. "Senior React Engineer", "Head of Sales")
• Location — city, metro, region, or remote (e.g. "New York City", "remote US")
• Skills / requirements — specific technologies, years of experience, domain expertise
• Seniority — junior, mid, senior, staff, director, VP, C-level
• Source preference — specific platforms (LinkedIn, GitHub, Indeed, etc.) or "all"

If a full job description was pasted, extract the above fields from it.

If role is missing or ambiguous, ask with AskUserQuestion:

"What role are you hiring for, and where? (e.g. 'Senior ML Engineer, remote US' or paste a job description)"

Once parameters are clear, confirm with the user using AskUserQuestion:

"Searching for: [Role] | Location: [Location] | Key skills: [Skills] | Seniority: [Seniority] Platforms to search: LinkedIn, Indeed, GitHub (for technical roles), AngelList / Wellfound, and professional communities. - Start search - Adjust parameters first"

Step 2: WSA Discovery

Discover available Web Search Agents for candidate-sourcing platforms. Run simultaneously:

nimble agent list --search "linkedin people" --limit 20
nimble agent list --search "indeed resume" --limit 20
nimble agent list --search "github profile" --limit 20
nimble agent list --search "wellfound talent" --limit 20

Filter results for entity_type: SERP or entity_type: PDP. Prefer managed_by: "nimble". Validate promising agents with:

nimble agent get --template-name {name}

Cache discovered WSA names and required params. If no WSAs found for a platform, fall back to nimble search for that platform.

Step 3: Parallel Candidate Search (Sub-Agents)

Spawn nimble-researcher agents (agents/nimble-researcher.md) with mode: "bypassPermissions", max 4 concurrent. Assign one agent per platform:

Agent 1 — LinkedIn

Search for people matching the role criteria. Use Boolean-style query construction:

nimble search --query "site:linkedin.com/in [Role] [Location] [Key Skills]" \
  --max-results 15 --search-depth fast
nimble search --query "[Role] [Location] linkedin profile [Skill1] [Skill2]" \
  --max-results 10 --search-depth fast

If a LinkedIn WSA was discovered in Step 2, use it instead with the role title, location, and skill keywords as inputs.

Agent 2 — Indeed / Resumes

nimble search --query "site:indeed.com resume [Role] [Location] [Key Skills]" \
  --max-results 10 --search-depth fast
nimble search --query "[Role] resume [Location] [Key Skills]" \
  --max-results 10 --search-depth fast

Agent 3 — GitHub (technical roles only)

Skip this agent for non-technical roles (e.g. Sales, Marketing, Operations).

nimble search --query "site:github.com [Role] [Location] [Key Skills]" \
  --max-results 10 --search-depth fast
nimble search --query "github [Key Skills] developer [Location] open to work" \
  --max-results 10 --search-depth fast

Agent 4 — AngelList / Wellfound + Communities

nimble search --query "site:wellfound.com [Role] [Location] [Key Skills]" \
  --max-results 10 --search-depth fast
nimble search --query "[Role] [Location] open to work OR seeking opportunities \
  [Key Skills]" --max-results 10 --search-depth fast

Each agent returns: candidate name (if available), profile URL, current title, location snippet, inferred skills, availability signals ("open to work", "seeking", "available") with event date (if available) and source URL.

Step 4: Deep Profile Extraction

For the top candidates identified in Step 3 (aim for 10–20 unique profiles across all platforms), extract full profile details. Run all extractions simultaneously:

nimble extract --url "[profile-url]" --format markdown

From each extracted profile, pull:

• Full name
• Current role & company
• Location
• Skills / tech stack
• Experience summary (years, notable employers)
• Education
• Availability signals (open to work, recent job change, posting activity)
• Contact signals (email, personal site, GitHub handle)

For extraction failures, follow the fallback pattern in references/nimble-playbook.md. If a profile is behind a login wall and extraction fails, keep the search-snippet summary instead — do not skip the candidate.

Extraction budget: extract up to 15 profiles. If more than 15 candidates were found in Step 3, prioritize by relevance score (seniority match + skill overlap + location match) before extracting.

Step 5: Score & Rank Candidates

Score each candidate (1–10) using these weighted signals:

Group candidates into tiers:

• Tier 1 (Strong match, 7–10): All required signals present
• Tier 2 (Partial match, 4–6): Most signals present, 1–2 gaps
• Tier 3 (Stretch, 1–3): Worth reviewing if Tier 1/2 list is thin

Step 6: Output

Before presenting results, check ~/.nimble/memory/talent-sourcing/[role-slug].md — if a candidate was surfaced in a prior run, mark them (previously surfaced) rather than re-presenting them as new.

Present a structured candidate report:

## Candidate Report: [Role] in [Location]
Searched: LinkedIn, Indeed, GitHub, Wellfound
Found: [N] candidates | Tier 1: [N] | Tier 2: [N] | Tier 3: [N]

**TL;DR:** [2-3 sentence summary of the strongest candidates and any notable patterns]

---

### Tier 1 — Strong Match

#### 1. [Name] — [Score]/10
- **Current role:** [Title] at [Company]
- **Location:** [Location]
- **Skills:** [Skill1], [Skill2], [Skill3]
- **Experience:** [X years, notable employers]
- **Availability:** [signal] — [event date or "date unknown"] — [source URL]
- **Profile:** [URL]
- **Contact signals:** [email / personal site / GitHub]

...

---

### What This Means
[1-2 sentences on hiring outlook: supply/demand signal, speed recommendation, any
standout sourcing channel]

Omit fields where data is unavailable. Do not fabricate details — use "unknown" for missing fields. Add a one-sentence "Why this candidate" note for each Tier 1 result.

Step 7: Save to Memory

Make all Write calls simultaneously:

• Report → ~/.nimble/memory/reports/talent-sourcing-{YYYY-MM-DD}.md (full candidate report with all tiers)
• Per-role → ~/.nimble/memory/talent-sourcing/[role-slug].md (candidate list; write or update)
• Profile → update last_runs.talent-sourcing in ~/.nimble/business-profile.json using the python3 snippet in references/profile-and-onboarding.md. Skip if the file does not exist.

Update ~/.nimble/memory/talent-sourcing/index.md with a row for this search. Follow the wiki update pattern from references/memory-and-distribution.md.

Step 8: Share & Distribute

Always offer distribution — do not skip this step. Follow references/memory-and-distribution.md for connector detection, sharing flow, and source links enforcement.

Step 9: Follow-ups

Offer next steps using AskUserQuestion:

What's next? - Go deeper on a candidate — extract full profile + find contact info - Expand search — broaden location, relax seniority, try more platforms - Narrow search — add a required skill or tighten location - Export list — save as CSV or formatted doc - Done

Sibling skill suggestions:

- Run company-deep-dive on a candidate's current employer for deal context - Run meeting-prep before reaching out to a Tier 1 candidate

Error Handling

See references/nimble-playbook.md for the standard error table. Skill-specific handling:

• Profile behind login wall: Keep search-snippet summary; note "full profile unavailable — LinkedIn/Indeed login required" in the candidate entry.
• < 5 total candidates found: Notify the user, suggest broadening location to remote or relaxing seniority, then ask whether to re-run with adjusted params.
• Search 500 on a platform: Retry once with a simplified query; if still failing, skip that platform and note it in the report header.
• GitHub agent skipped for non-technical role: Note "GitHub not searched for this role type" in the report header.