conjurer

name	conjurer
description	This skill should be used when the user asks to "tailor my resume for this job", "help me apply to this role", "draft a tailored resume and cover letter for this job posting", "generate resume/cover letter variants", "conjure variants for this role", or "run the conjurer pipeline". Generates grounded resume and cover letter variants from a master resume and a grimoire of encoded taste, curates them conversationally, and stitches final documents.
version	0.1.0

Conjurer

Generate many grounded variants of resume bullets and cover letter paragraphs, curate the best with the user, and stitch them into final documents. Every claim is grounded in the user's master resume or evidence; the grimoire encodes voice and taste. Generate abundance, exercise taste, stitch from selections.

Scripts live in scripts/ (run with bare python3, no dependencies). Detailed procedure is in references/pipeline.md. Generation uses the variant-generator subagent.

Workspace: default to the current directory

A workspace is a directory containing grimoire.md, master-resume.md, and applications/. By default the workspace is the directory the user launched Claude Code in (the current working directory). Resolve it before generating:

• Workspace already here — cwd contains grimoire.md and master-resume.md: use it. Go straight to the pipeline.
• First run — no grimoire.md/master-resume.md in cwd: bootstrap one here.
  1. Ask whether the user has an existing resume to start from. Accept a file path or pasted text and read it: for a PDF use the Read tool directly; for a .docx run python3 <SKILL_DIR>/scripts/extract_text.py <file.docx>; for markdown or plain text read it directly.
  2. Run python3 <SKILL_DIR>/scripts/new_workspace.py . to scaffold any MISSING workspace files (grimoire.md, master-resume.md, applications/) into the current directory. It does not clobber files that already exist, so running it in a directory that already holds the user's resume is safe.
  3. Build the master resume: ingest the user's existing resume into master-resume.md (the /conjurer:master-resume flow). If they have no resume, interview them.
  4. Build the grimoire: run the /conjurer:grimoire flow (interview, optionally ingesting the same resume and any writing samples).
• Override — if the user keeps their grimoire or master resume elsewhere (for example files they already maintain in another repo or vault), accept explicit paths and pass those to the steps below instead of the cwd copies.

<SKILL_DIR> is this skill's directory. Do not generate against unedited templates — confirm grimoire.md and master-resume.md carry the user's real content first. A grimoire.md still containing the FILL THIS IN marker, or a master-resume.md still naming the placeholder companies Acme Corp/Globex, has not been filled in; run the relevant builder command before generating.

/conjurer:grimoire and /conjurer:master-resume are commands provided by this same plugin, so they are available whenever the conjurer plugin is installed.

Pipeline

1. Ensure workspace (above). Confirm grimoire.md and master-resume.md are filled in, not templates.
2. Init the application: python3 <SKILL_DIR>/scripts/init_app.py <slug> <workspace>. Ask the user for the job description (paste into applications/<slug>/jd.txt) and any extra evidence (evidence.md).
3. Outline: read the grimoire, master resume, JD, and evidence. Choose one strategic frame and design the unit skeleton. Write applications/<slug>/outline.json. See references/pipeline.md for the exact schema and the four frames.
4. Variants: for each unit in the outline, dispatch the variant-generator subagent, passing the grimoire, master resume, evidence, outline, that unit, and N = 4 variants per unit (the agent requires a count). Use the Task tool with subagent_type: variant-generator, one call per unit, and issue all unit calls in a single tool-use block so they run in parallel. Assemble the returned blocks into applications/<slug>/variants.md. See references/pipeline.md for the file format.
5. Curate: present the variants conversationally. When the user picks, mark exactly one - [x] Pick per unit in variants.md.
6. Stitch: python3 <SKILL_DIR>/scripts/stitch.py applications/<slug> <workspace>/master-resume.md. Writes cover_letter.md and resume.md.
7. Lint: python3 <SKILL_DIR>/scripts/lint.py applications/<slug>. Report findings; fix any and re-run until clean.
8. Export (optional): offer to export the finished documents for submission with python3 <SKILL_DIR>/scripts/export_docs.py applications/<slug> pdf docx. This uses pandoc if it is installed; if not, it reports that the markdown sources are the deliverable. Tell the user which outputs were written or skipped.

Claim discipline

Never invent claims, numbers, technologies, or outcomes. Every variant re-frames a fact that exists in the master resume or evidence, and cites its source. If the user asks for a claim that is not in the evidence, say it is unsupported rather than fabricating it.

Additional Resources

• references/pipeline.md — outline schema, strategic frames, variants.md format, exact commands.
• scripts/ — new_workspace.py, init_app.py, stitch.py, lint.py, extract_text.py (docx ingest), export_docs.py (PDF/docx export).
• agents/variant-generator.md (plugin agent) — the per-unit generator.
• Workspace setup: the /conjurer:grimoire and /conjurer:master-resume commands.