CareerRat Claude Instructions

Follow AGENTS.md. It is the source of truth for intent routing, the body-read gate, artifact completeness, and submit safety. If candidate/AGENTS.md exists, read it too for candidate-specific context; if the workspace is not set up yet, run ingest-profile (or careerrat ingest) first.

Key rule: when the user asks to apply, run the apply-job workflow. apply-job must run or verify evaluate-job before tailoring, filling, or submitting.

When the user asks to draft, reply to, follow up on, summarize, or negotiate a job-search email or recruiter message, run email-comms and update tracker communication state.

When the user asks to schedule, propose times for, confirm, or reschedule a recruiter or hiring-team call or interview, run schedule-meeting (it owns timezone resolution, double-booking avoidance, calendar-ready holds, and scheduling write-back; non-scheduling threads hand back to email-comms).

When the user is in a live or verbal offer or negotiation call, or wants to rehearse or script a verbal comp conversation, run interview-prep (live negotiation channel).

When the user asks to optimize, review, or improve their LinkedIn profile, or to make it read for the roles they're targeting, run optimize-linkedin. It produces a read-only before→after diff first (the default), and only writes edits back when profile_apply is enabled, confirm-first per field. In the app, suggestions are staged automatically with an Undo, and writing to LinkedIn still asks per field.

When the user asks how risky, stable, or healthy a company is (layoffs, finances, morale, "is this a safe place to land"), or to factor company risk into a role, run company-health. It scores a role-scoped healthy|watch|risky rating with provenance, persists it to the tracker, and feeds the fit score only where it cross-cuts a stated candidate need. It runs on its own when an application reaches the fire stage (interview by default, modes.yml#company_health) with no fresh rating, and on demand anytime; the rating is an internal signal and never enters an outbound artifact.

When a careerrat command crashes or behaves like it's broken, or the user asks to report a bug or file an issue, run report-issue. It separates a real defect from a config problem, assembles redacted diagnostics (never candidate PII, comp, or workspace contents), and only with an explicit yes opens a GitHub issue on the upstream CodesWhat/careerrat repo. It never auto-files.

When the user pastes an application-form or screening question and wants a drafted answer outside a full tailor/apply run ("how should I answer this", "the form is asking X"), run answer-question. It reuses screening_answers first, grounds new answers in profile/honesty/evidence, never fabricates (the unanswerable case is the literal NEEDS YOU marker), and persists durable disclosure-style answers back to form-defaults.yml#screening_answers so they're never re-asked. During a full tailor/apply run, tailor-application STEP 6 owns the answers artifact instead.

When the user asks how to close a fit gap, improve their fit for a role, or what to do about a job that landed at review, run coach-gaps. It only fires on an evaluate-job verdict already at gate: review with named fitRisks, never on keep or cut. It runs on its own when that verdict lands (once per evaluation), and on demand from the Coach me button. Every suggestion is grounded only in evidence already on record or the current conversation — an honest "no close path yet" is a valid, expected answer for a gap, not a failure. A saved evidence-claim draft routes through the same evidence firewall (computeEvidenceWrite/candidateEvidenceMerge) every other evidence write uses; evaluate-job itself stays the narrow, unchanged gate.