talent-scout

Single-candidate evaluation — evaluates a pre-enriched candidate against the user's thesis, produces dimension ratings, assigns a tier, and persists a ScoreRecord. Requires a candidate meeting the OSINT minimum from the profile. Dispatched by scout, manage, or orchestrator skills. Examples: <example> Context: Orchestrator enriched a batch of candidates. Now they need scoring. user: "go" assistant: "Dispatching 8 assess subagents in parallel to score the enriched batch." <commentary> After enrich agents populate OSINT data, assess agents score each candidate independently against the thesis. </commentary> </example>

Single-candidate deep investigation — synthesizes enriched OSINT data into a structured report with investment thesis, bear case, and connection analysis. Requires a candidate meeting the deep-dive OSINT minimum from the profile. Dispatched by the deep-dive skill or orchestrator. Examples: <example> Context: Orchestrator fully enriched a top candidate. Now needs full investigation. user: "go" assistant: "Dispatching deep-dive subagent to synthesize report and thesis." <commentary> After enrich agent fills OSINT sources, deep-dive agent synthesizes the data into a report grounded in the user's thesis. </commentary> </example>

OSINT enrichment agent — runs WebSearch for deep-dive-level sources on a shortlisted candidate, structures findings, and persists via pipeline tools. Dispatched only after a candidate has been promoted to deep-dive. Examples: <example> Context: User selected 2 candidates for deep-dive after a scout pass. user: "go" assistant: "Dispatching 2 enrich subagents to fill OSINT sources before deep-dive synthesis." <commentary> Enrich runs only in deep-dive mode now. Scout-level evidence comes from the LinkedIn profile loaded by ingest_candidate. </commentary> </example>

Network expansion from known high-value candidates — discovers similar profiles via LinkdAPI, extracts and saves new candidates to the pipeline. Dispatched by the expand skill or scout/orchestrator. Examples: <example> Context: Orchestrator has deep-dived a top-tier candidate and wants to mine their network for similar profiles. user: "go" assistant: "Dispatching expand subagent for ron-shemesh to discover similar profiles in his network." <commentary> After deep-diving a high-conviction candidate, orchestrator dispatches expand to find new candidates through network proximity. The caller will dispatch assess agents for any new IDs returned. </commentary> </example> <example> Context: Scout wants to broaden discovery by expanding from 3 high-value candidates found in the current session. user: "go" assistant: "Dispatching expand subagent for network discovery from ron-shemesh, eli-cohen, and noam-nissan." <commentary> Scout dispatches expand with multiple seed candidate IDs. Expand discovers and saves new candidates, then returns their IDs so the caller can dispatch assess agents. </commentary> </example>

Company intelligence agent — discovers companies matching the user's thesis via WebSearch and LinkdAPI, verifies and persists them to the pipeline. Dispatched by orchestrator during company refresh. Examples: <example> Context: Orchestrator detects companies.json is stale (last refresh past the refresh interval). user: "run the pipeline" assistant: "Dispatching research-companies agent to discover new acquisitions and companies." <commentary> The orchestrator dispatches this agent during company_refresh when company data is stale. </commentary> </example> <example> Context: First session after init — companies.json exists but has never been refreshed. user: "what's next" assistant: "Company data has never been refreshed. Dispatching research-companies agent before discovery." <commentary> On first session, companies.json has last_full_refresh: null, triggering company_refresh. </commentary> </example>

Use for status checks ('what's the status', 'debrief') or as session-end procedure (called by orchestrator). In standalone mode, loads system state and summarizes pipeline — does not execute any pipeline work. In session-end mode, receives an action log from orchestrator, writes session log, and commits to git. No longer the default session-start skill — orchestrator handles that when the user says 'go'.

Use when the user asks to deep-dive, investigate, or look more carefully at a specific candidate by name. Triggers on 'deep-dive <name>', 'deep dive <name>', 'investigate <name>', 'look more carefully at <name>', 'tell me more about <name>', 'who is <name>'. Resolves the name to a candidate id, ensures enrichment, dispatches the deep-dive agent. Does NOT trigger for general session work ('start scouting', 'continue', 'what's the status') — those go to orchestrator/debrief.

Health check for talent-scout installation. Verifies dependencies, MCP servers, data directory, and API connectivity. Use when something seems broken, after installation, or when the user says "doctor", "health check", "diagnose", "what's wrong", "verify setup", or "is everything working".

Use when the user asks to expand from a specific candidate's network. Triggers on 'expand from <name>', 'expand <name>', 'find similar to <name>', 'people like <name>', "who's in <name>'s network", 'mine <name>'s connections'. Resolves the name to a candidate id, dispatches the expand agent, then enriches and assesses any newly discovered candidates. Does NOT trigger for general session work — those go to orchestrator.

First-time setup and onboarding for talent-scout. Triggers automatically via SessionStart hook when ~/.talent-scout does not exist. Captures the LinkdAPI key, delegates thesis drafting to the `ts-interview` skill, then atomically writes all config on user confirmation. Also use when the user says "reconfigure talent-scout", "reset setup", or "reinitialize".

1 hook across 1 event