Claude Resume Kit for Codex

This is a Codex-native wrapper around the upstream ARPeeketi/claude-resume-kit project. The upstream project is preserved under assets/upstream-project/; treat it as bundled reference material and reusable scaffold, not as already configured user data.

Source And Provenance

• Upstream repo: https://github.com/ARPeeketi/claude-resume-kit
• Installed upstream commit: 69930e9de21d5595f9b8c9427adc9c51a9fcbc0e
• License: MIT, preserved at assets/upstream-project/LICENSE
• Original Claude slash skills live in assets/upstream-project/.claude/skills/

Codex Adaptation

The upstream workflow is slash-command based. In Codex, route natural-language requests to the equivalent upstream sub-workflow:

User intent	Upstream file to read
Extract a paper, report, PDF, .tex, or codebase into resume evidence	assets/upstream-project/.claude/skills/setup-extract/SKILL.md
Build or refresh the resume knowledge base from extractions	assets/upstream-project/.claude/skills/setup-build-kb/SKILL.md
Generate a tailored resume or academic CV from a JD/opportunity	assets/upstream-project/.claude/skills/make-resume/SKILL.md
Generate a matching cover letter	assets/upstream-project/.claude/skills/make-cl/SKILL.md
Critique a generated resume/CV/cover letter package	assets/upstream-project/.claude/skills/critique/SKILL.md
Apply critique or user feedback to a package	assets/upstream-project/.claude/skills/edit-resume/SKILL.md

Read only the relevant upstream sub-skill and the referenced upstream files it names. Do not load the entire upstream project unless the task really needs broad migration or debugging.

Workflow

1. Identify whether the user is asking for extraction, knowledge-base setup, resume/CV generation, cover-letter generation, critique, or edit.
2. If starting a new project, create or use a working directory chosen by the user. Copy the scaffold from assets/upstream-project/ into that working directory, excluding .git and any generated private output the user does not want copied.
3. Preserve the upstream directory contract inside the working project:
   • config.md
   • CLAUDE.md
   • knowledge_base/papers/
   • knowledge_base/extractions/
   • resume_builder/experience/
   • resume_builder/bundles/
   • resume_builder/templates/
   • JDs/
   • output/
4. Follow the selected upstream sub-skill, translating slash commands into ordinary Codex steps.
5. Preserve mandatory user-confirmation stops from upstream instructions. When the upstream workflow says to stop and wait, stop and ask the user rather than continuing silently.
6. Use the bundled LaTeX templates and helper scripts when generating final .tex outputs. If a local TeX toolchain is unavailable, still produce Markdown/session artifacts and clearly state that PDF compilation was not verified.

Accuracy Rules

• Never claim unpublished, under-review, internal, draft, or coursework material is published.
• Never upgrade contribution verbs beyond evidence: use full-ownership verbs only for work the user actually led or solely performed.
• Never invent author position, publication venue, DOI, acceptance status, metrics, awards, affiliations, dates, roles, or tools.
• Track provenance for every achievement: source artifact, publication status, user's contribution, safe claims, hedged claims, and do-not-claim items.
• If evidence is missing, ask targeted questions or mark the item as uncertain.

Practical Use In Codex

For a paper-to-CV task:

1. Read setup-extract/SKILL.md.
2. Extract metadata, methods, results, collaboration scope, user contribution, provenance notes, and bullet seeds.
3. Write the extraction into the working project's knowledge_base/extractions/.
4. Update _INVENTORY.md if present, or create it if missing.

For a tailored resume/CV task:

1. Read make-resume/SKILL.md.
2. Read the working project's config.md, selected JD/opportunity, relevant experience files, role bundle, and resume references.
3. Produce a session file, bullet plan, and then the resume/CV only after user confirmation where required.

For critique:

1. Read critique/SKILL.md.
2. Evaluate factual integrity, provenance, fit, formatting, AI-sounding language, and reviewer-perspective risk.
3. Return prioritized fixes and update the session file if the working project uses one.

Boundaries

• Do not store secrets, tokens, private account data, cookies, or raw private transcripts in the skill or generated reusable scaffold.
• Do not edit the preserved upstream files inside assets/upstream-project/ unless explicitly updating the installed skill itself.
• Do not treat the upstream example persona as user data.
• Do not submit applications, message recruiters, or automate platform actions.