Apply — job-scoped application materials, end to end

Goal

A submitted application the candidate is proud of. An application ships THREE artifacts: tailored résumé (+ ATS PDF), cover letter, and outreach plan (contacts/<company>.md with recruiter/HM message drafts). A cold submit without outreach is half an application.

Prerequisites

• Required: Workspace CLAUDE.md (none: route to profile's Setup); base-resume.md (SSOT: body + § Claim rules — hard stop if absent, never fabricate); jd-analysis/ for the role (suggest evaluate first if absent).
• Optional: profile.md (+ § Application defaults), storybank.md, pitch.md, voice.md.

Loops and sequences

Extended craft and form automation in references/patterns.md. File shapes in references/schema.md.

Tailoring (the loop)

Runs when the candidate asks to tailor a résumé, letter, or "materials" for a specific role.

• Standard: The ## Standard bullets in the application file (references/schema.md).
• Budget: Two self-passes; one review incorporation round, which re-checks only the lenses short of Pass, once. Said up front.
• Each round: Read ## Standard and earlier ## Rounds rows first. Reshape-only from base-resume.md (selection, order, depth — never invented facts). Run the checks and the three-lens panel (references/eval.md § The panel — three lenses). Score round as N/M held count and append row to ## Rounds.
• Obligations:
  1. Facts need a yes: Voluntarily surfaced facts go to storybank.md, not directly into a résumé without an explicit yes.
  2. Deliver & Disclose: Present the deliverable beside the proposal block (scripts/proposal_block.mjs list of cuts and placed words with easy reversal).
  3. Exact edits: Apply candidate edits directly in place and re-render.
• Exits: When the round scores M/M held with every lens at Pass and the candidate accepts; at the budget; or at the ceiling: two ## Rounds rows with the same count and lens verdicts. At the budget or the ceiling, stop and present a decision for the candidate to make, showing each lens still at Revise or Fail with its ## Panel rows; the package is not ready until every lens passes or the candidate waives it in chat. Never relax the standard — cutting a claim's supporting evidence to fit a limit IS relaxing it: a claim-name without its number is not the claim.

Answers (the loop)

Runs when application questions or form fields need drafting. Follow references/patterns.md § Application answers and voice.md.

• Standard: Answer addresses the exact prompt within word limits; CONFIRMED stories only.
• Budget: 2 rounds; ceiling applies.
• Exits: All answers in ## Answers with gaps flagged; or at the ceiling with tradeoff surfaced. Never relax the standard — cutting a claim's supporting evidence to fit a limit IS relaxing it: a claim-name without its number is not the claim.

Full application (a sequence)

Runs when candidate says "apply to this one".

1. Extract & Draft: Extract all form fields/pages into the application file before drafting essays. Screening answers pull from profile.md § Application defaults.
2. Review Gate: Complete application (fields, essays, chosen résumé) approved in chat before touching the live form, with any LinkedIn drift named.
3. Submit Gate (Rule 7): Form filled in the browser, then a screenshot and a plain summary, then the candidate's explicit word fires submit.
4. Third Artifact: Capture submission confirmation into ## Submission, set jobs.md row to Applied, and draft the outreach plan in contacts/<company>.md (## Outreach plan).

Exits with all three artifacts delivered or at the gate candidate declined.

Batch prep (a sequence)

Runs when "prepare top N" (default N ≤ 5). User-armed, never scheduled.

• Ensure each job has an evaluate decode; run the tailoring loop for résumé/letter; draft form questions; build outreach plans. Never auto-submit in batch.

Exits with one status line per job.

State

Owned: applications/ (shapes in references/schema.md).

• Hands back: Outreach sends go to outreach; interview prep goes to interview; confirmed story leads go to storybank; base updates go to profile.
• Session close: Run node scripts/check_materials.mjs, node scripts/proposal_block.mjs, and node ../profile/scripts/check_files.mjs --workspace .. Wording check subagent runs on every delivered document. Report outcomes, never narration (clean is 1 line, and only when nothing failed or warned; fix FAILs before the reply ends; fix each WARN or tell it to the candidate).

Guardrails

• Reshape-only, never invent: No claim the base doesn't support. Unsupported sections are omitted, never filled with filler.
• Never: Create accounts, handle passwords, bypass CAPTCHAs, guess screening answers, or claim "submitted" without captured confirmation.

Every reply ends with ONE contextual next step — a sentence with its why, not a menu.