exec-recruiter

Sources senior product roles directly from major ATS domains (Greenhouse, Lever, Workday, Ashby, iCIMS, Jobvite, BambooHR, SmartRecruiters, Workable, Teamtailor) via Google's site: operator. Catches roles that never reach LinkedIn or arrive there days late. Used by the phase-1-sweep skill as Step 1B alongside the LinkedIn sweep. Returns a structured list of live, on-ICP candidate URLs ready to feed into the existing scoring pipeline. <example> Context: Phase 1 daily sweep needs to source roles from beyond LinkedIn. user: "Run the ATS-direct sweep for UK and US-remote VP Product roles." assistant: I'll build per-region site: queries against the 10 ATSs, run liveness checks on every URL before fetching, route SPA-based ATSs (Ashby, Workday) through the jd-extractor for Chrome render, and return a deduped candidate list. <commentary> ATS-direct sourcing has different failure modes than LinkedIn (staleness, SPA shells, recruiter-aggregator pollution). The agent handles each. </commentary> </example> <example> Context: A specific company isn't on LinkedIn but might post on its ATS. user: "Did Front, HiveMQ, or Compyl post a senior product role recently?" assistant: I'll run targeted site: queries against jobs.lever.co and jobs.ashbyhq.com filtered by company name. Liveness check first, then fetch only live URLs. <commentary> The agent is also useful for ad-hoc, company-targeted ATS lookups, not just batch sweeps. </commentary> </example>

Applies JD-driven tailoring to a baseline CV docx using a strict discipline. Identity blocks (name, headline scaffold, dates, education) stay stable. Tailorable fields are limited to one headline-positioning fragment, one adjacency sentence at the end of the executive summary, at most two of six career-highlight cells, the italic context line per relevant past role, and the geo-conditional header. Generates the tailored docx via XML-level text replacement (no python-docx required) and converts to PDF via libreoffice. Use this whenever a CV needs to be tailored to a specific role - never rewrite the CV from scratch. <example> Context: Phase 2 packet build needs a JD-tailored CV for a specific role. user: "Tailor the CV for brevia-vp-product" assistant: I'll apply the discipline rules: change only the headline-positioning, the adjacency sentence, one or two highlight cells, and the italic context lines per role. The rest of the CV stays exactly as the baseline. <commentary> The discipline preserves identity. Without it, every tailored CV becomes a one-off rewrite, which the user has explicitly flagged as unacceptable. </commentary> </example> <example> Context: Mid-process JD reveal changes the role's framing significantly. user: "The actual JD reads as Office-of-the-CPO, not line product. Refresh the CV." assistant: Same discipline applies - I'll swap only the headline, adjacency, one highlight cell, and italic context lines, this time around the IC framing. Identity stable. <commentary> Reframing happens in the discipline-permitted fields, not by rewriting the resume. </commentary> </example>

Pulls full job description bodies via the LinkedIn-to-ATS redirect URL extraction trick. LinkedIn's `/jobs/view/` page renders a skeleton in unauthenticated/limited views, but the actual JD lives behind the "Apply on company website" link, which is a `/safety/go/?url=...` redirect to the underlying ATS. This agent recovers the destination ATS URL, navigates there, and extracts the full JD body for use in scoring, CV tailoring, and outreach drafting. <example> Context: Phase 1 sweep needs the full body of a LinkedIn job whose card-level data alone is insufficient for scoring. user: "Pull the full JD for https://www.linkedin.com/jobs/view/EXAMPLE001/" assistant: I'll extract the destination ATS URL from the apply-link redirect, navigate, and pull the full JD body. <commentary> LinkedIn unauth view is a skeleton; the JD body is at the destination ATS. Use the redirect-URL extraction trick to recover it. </commentary> </example> <example> Context: Phase 2 packet build requires a JD body to derive replacements for CV tailoring and to identify anti-bot screener words. user: "Build the packet for brevia-vp-product" assistant: First I'll pull the full JD via the ATS redirect so I can tailor the CV against the actual mandate signals and detect any screener words like 'crane'. <commentary> Card-level data misclassified the title implication; the body has the real reporting line, customer names, and screener test. </commentary> </example> <example> Context: Score-role skill receives a LinkedIn URL with no other context. user: "score this https://www.linkedin.com/jobs/view/EXAMPLE002/" assistant: I'll pull the full JD body first - the title-level read at LinkedIn often contradicts the body (saw this happen with [sample-healthcare-AI-co]'s "Head of Product Transformation" being a chief-of-staff IC role). <commentary> Always extract the body before scoring. Card-only scores are low-confidence. </commentary> </example>

Research a named interviewer (LinkedIn + TheOrg + Crunchbase + web search) and refresh company intel for a meeting prep packet. Used by phase-4-prep skill. Produces structured interviewer profile + company-intel-refresh notes that phase-4-prep populates into the meeting prep templates.

Drafts outreach messages (Connect+Notes, longer DMs, InMails, application Summary text, cover letters) per the user's outreach style discipline. Applies the channel character limits, peer-to-peer voice, banned punctuation/phrases, required structural elements, and JD-specific hooks. Always produces the variants the calling skill needs (primary contact + backup contact, short Connect+Note + longer DM, sometimes a cover-letter Summary). Never sends - returns the drafts for the calling skill to surface to the user. <example> Context: Phase 2 packet build needs Connect+Notes for both the CTO and CEO of a target company. user: "Draft outreach for brevia-vp-product, primary Alan Mosca (CTO), backup Dev Amratia (CEO)" assistant: I'll draft two Connect+Notes (300 char Premium each), one for Alan with the CTO/AI-native angle and one for Dev with the founder-level angle, plus a longer DM for use after a connection accepts. <commentary> Always produce both primary and backup variants. Always include a longer DM for the post-accept conversation. </commentary> </example> <example> Context: A workable application form has an optional Summary box and a JD-mandated screener word ("crane"). user: "Draft the Summary for the [sample-construction-AI-co] Workable application; the JD requires the word 'crane' to be present." assistant: I'll draft a 1500-char Summary that lands the JD-specific hooks, embeds 'crane' naturally in a sentence about the user's pattern match to the AI-for-construction thesis. <commentary> Application form copy is just another channel. Same discipline rules apply (no em/en dashes, peer voice, no banned phrases). Anti-bot screener words go in naturally, not flagged as a screener-test inclusion. </commentary> </example>

Generate Day-5/7/10 follow-up bump messages for any pending roles. Triggers on "what bumps today", "Day-5 bumps", "follow-ups due", "/bump", "/bump X", or "what should I follow up on today". Reads the dashboard, finds roles in the bump window or past Day-10 final-action threshold, and drafts the appropriate bump message per role per follow-up.yaml templates.

Read the current dashboard and surface today's actions, status counts, and follow-up windows. Triggers on "where do I stand", "today's actions", "dashboard status", "/dashboard", "what's the state", "status check", or first thing in the morning before the sweep. Lightweight read-only operation that summarizes the pipeline without changing anything.

Run the daily Phase 1 sweep for senior product leadership roles. Triggers on "run today's sweep", "Phase 1", "find new roles", "/sweep", "what's new today", "morning sweep", "scan LinkedIn for product roles". Sources roles from LinkedIn AND directly from major ATS domains (Greenhouse, Lever, Workday, Ashby, etc.) via Google site: search, pulls JD bodies via the apply-link redirect or direct fetch, scores each role 1-10 against the user's ICP and scoring weights, dedupes against the existing dashboard, and produces a Tier-1 / Tier-2 / Skip greenlight summary. Stops before Phase 2 - waits for the user's per-role greenlight.

Build a Phase 2 outreach packet for a single greenlit role. Triggers on "build packet for X", "tailor CV for X", "Phase 2", "packet X", "/packet", or any request to prepare an application for a specific company / role. Pulls the full JD via the ATS redirect, identifies the right contacts, generates a JD-tailored CV (docx + PDF) per the discipline rules, drafts channel-appropriate outreach (Connect+Note + longer DM), and saves the packet under outreach/{slug}/. Stops before Phase 3 - waits for the user's explicit "send" approval.

Walk through Phase 3 send for a single greenlit packet. Triggers on "send X", "Phase 3 X", "/send X", "fire X", "let's send X", or "ready to send X". Opens the right LinkedIn surface for each channel (Connect+Note dialog, DM thread, InMail compose), types the drafted message into the field, and stops with the cursor positioned for the user to click Send. Never sends autonomously. Walks the user through each channel in priority order; after each user-confirmed send, updates the dashboard from "drafted" to "pending" with sendDate today.