Conference Recruiting

Conference Recruiting Sweep

Conference sponsorship signals budget and presence; speaker, village, booth, and CTF affiliations signal active technical teams. Neither signal proves candidate fit. The sweep turns those signals into a verified, human-route-first pipeline.

Signal Model

Core Rules

1. Require a configured profile source; never guess. Resolve PROFILE SOURCE only from an explicit path in the current request or from CONFERENCE_RECRUITING_PROFILE_SOURCE (explicit path wins). There is no default. Before any fit judgment or worker dispatch, verify that the resolved path is a readable regular file, read it, and validate it against the profile contract. If it is unset, unreadable, or invalid, stop and ask the user to configure or repair it. BrainLayer may supplement preferences and pipeline history, but experience, positioning, artifact, connector, and disclosure claims come only from the profile source. Every worker brief quotes the relevant block verbatim with exact file+line provenance using the worker template.
2. Mine sponsors and speakers. Resolve the official conference source, extract sponsors/exhibitors with tiers, and extract speaker affiliations; dedupe while retaining every source tag. Speaker mining may be omitted only as an explicit, reasoned opt-out recorded before the final report. Sponsor volume is never a silent opt-out. See conference sources.
3. Verify live hiring from first parties. For JavaScript-heavy boards, try the official ATS API before mirrors. A cached or aggregator listing is discovery, not proof. Read the live posting body and direct apply URL. If verification fails after fallback search, report could not check, never not hiring.
4. Route by actual level. Unprefixed roles are cold-applyable. Senior-titled roles are referral-only; cold Senior is dead. A Senior / Principal slug is a dual-level req, not one title: read the body for the real level and clarify a referrer's misread with quoted body evidence. Uniform Senior prefixes are a level floor, not flexibility created by req volume. Apply job-routing doctrine.
5. Rank by whether a live human exists, and sort by it. Cross-check past applications, rejections, referrals, and whatever contact evidence actually exists before ranking. A connector roster is not proof of a live opportunity: unknown is the default, and only current profile-source evidence, a separately verified current record, or user confirmation may promote it. Never infer unworked or active state from a person's presence in the roster. The report must contain an Is there a human here? column and sort verified active paths first. A referral changes who reads the candidate, never what level the candidate is.
6. Choose resumes from the configured artifact inventory. Read the validated profile source's artifacts records. Shipment folder names are not variant identities; rendered headline strings are. Before a send, verify that the selected artifact path is a readable regular file. Send one resume per referrer human even for multiple roles. Per-role variants are allowed only across different recipients. Headline matching beats body tailoring, and no new tailoring build is the default. Repairing an untruthful headline is a defect fix, not a tailoring build, and is always allowed: block the send and route the repair/rebuild through the workflow that owns the profile source. If artifacts is empty or the selected path cannot be read, stop and ask the user; never re-derive inventory from folder names.
7. Protect the referrer. Before working a new lead, add the connector through the workflow that maintains the profile source and re-read the regenerated or updated source. Never edit a generated export directly. Send the chosen CV with the exact req link and an upfront, truthful years-gap disclosure. Make one concrete ask per human, with secondary options clearly secondary. A fast disclosed no is a cheap successful outcome, not a protocol failure. Follow the full referral protocol.
8. Bound and prove fan-out. Pin a concrete model ID for every worker, verify the effective runtime model, and stagger waves at a maximum of three concurrent workers. Any eval record must separately capture requested model, observed effective model ID, observed effective effort, and an allowed runtime observation source. An alias or model-table inference is not eval evidence.
9. Ship one evidence-backed report. Date-stamp it; sort by human route first and location second; include role title, body-confirmed level, location, source signal, pipeline state, human state, direct URL, and verification route. Keep watch and could-not-check sections. Never fabricate a role. Write the primary report, copy it only to explicitly configured secondary destinations, shortlist 5–10 targets, and brain-store the summary. Follow the report contract.

Run

1. Resolve and validate PROFILE SOURCE, fill the worker template from it, then gather conference and live-job evidence in bounded, model-verified waves.
2. Apply job routing and pipeline history before ranking. For warm routes, read the configured artifact inventory and referral protocol before choosing the package.
3. Produce, copy, shortlist, and store the report exactly as the report contract specifies.

Boundaries

• The profile-source owner owns resume artifacts, shipment paths, headline truth, and builder edits. This skill consumes the configured contract rather than copying volatile facts. It routes truth repairs through the owning workflow; the no-build default never permits a known false headline to keep shipping.
• General recruiting stays with the user's configured job-search workflow; this skill produces a conference-sourced input to that pipeline.
• /never-fabricate applies to every role, model-provenance claim, and completion claim.