Communitygithub.com

ncrmro/.agents

Human-invoked. Create and maintain a canonical personal resume record, produce role-specific resume variants, track work history and evidence, and synchronize the public resume website. Use when explicitly asked to create, tailor, audit, export, or publish a resume or CV.

¿Qué es .agents?

.agents is a Claude Code agent skill that human-invoked. Create and maintain a canonical personal resume record, produce role-specific resume variants, track work history and evidence, and synchronize the public resume website. Use when explicitly asked to create, tailor, audit, export, or publish a resume or CV.

Compatible con~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/ncrmro/.agents/tree/HEAD/skills/resume-builder

Preguntar en tu IA favorita

Abre un nuevo chat con esta habilidad de agente ya precargada.

Documentación

Resume builder

Maintain one evidence-backed career record under docs/personal/resumes/, derive targeted resumes from it, and publish only the explicitly public subset to a personal website.

Read these supporting docs when needed:

  • references/convention.md — required file layout, record fields, claim IDs, variants, and publication state.
  • references/website-sync.md — discovering and updating a resume website without making it a second source of truth.
  • references/source-recovery.md — recovering work history from legacy sites, repository history, bookmark managers, and archived resumes.
  • assets/scaffold/ — starter files for an empty resume root.

Default decision

Facts live once; variants select them by stable ID. Never copy a role or achievement into a variant and let it drift. Update the canonical record, then select, order, or narrowly reframe its claims for a target role. A website is a publication target, not the canonical work-history database.

Safety and truthfulness

  • Never invent dates, titles, metrics, credentials, employers, clients, or technologies. Mark uncertain material status: needs-verification.
  • Preserve the most precise date actually known; do not manufacture a day to satisfy a format.
  • Treat contact details, compensation, references, customer names, and non-public evidence as private unless explicitly marked otherwise.
  • Do not publish or commit private fields to a public repository.
  • Keep evidence beside the canonical claim as a link or short provenance note; do not copy confidential source material into the resume tree.
  • Show the user material wording changes and unresolved facts before publishing.

Workflow

  1. Locate the personal source.
    • Read the governing AGENTS.md.
    • Use the repository root containing docs/personal/resumes/.
    • If the directory is absent, offer to copy assets/scaffold/ there.
    • Read its README.md and all records involved in the requested change.
  2. Inventory before editing.
    • List canonical roles, projects, claim IDs, variants, and publication targets.
    • Compare the record with the current resume website, repository history, bookmarked sources, and any supplied resume files.
    • Report contradictions, stale open-ended roles, and unverified claims.
    • For a legacy bootstrap, follow references/source-recovery.md.
  3. Update canonical facts first.
    • Add or amend work-history/ and projects/ records.
    • Give each reusable achievement a stable claim ID.
    • Record visibility, confidence, evidence, and last_verified.
    • Close prior open-ended roles when a newer fact establishes an end date.
  4. Build or refresh variants.
    • Create variants/<slug>.md from the canonical profile and selected claim IDs.
    • Tailor summary, ordering, and emphasis to the target role.
    • Keep job-description keywords only where the canonical evidence supports them.
    • Link application-specific variants to the existing application record rather than duplicating application tracking here.
  5. Render and review.
    • Resolve every selected claim ID.
    • Check chronology, tense, consistency, page length, contact fields, links, and spelling.
    • Present material edits and any needs-verification items to the user.
  6. Synchronize publication targets.
    • Follow references/website-sync.md.
    • Update only records with visibility: public.
    • Preserve the target repository's content schema and local instructions.
    • Build or run the narrowest available validation before publishing.
  7. Record freshness.
    • Update the variant's last_reviewed.
    • After a verified website update, update the target record's last_published and source revision or commit.
    • Report what remains unpublished or unverified.

Variant rules

A variant may:

  • choose and order canonical claim IDs;
  • use a target-specific summary;
  • shorten a supported claim without changing its meaning;
  • group supported skills for readability;
  • omit irrelevant roles, projects, and claims.

A variant must not:

  • create facts that exist only in the variant;
  • inflate scope, seniority, ownership, or numerical impact;
  • expose a private claim;
  • silently change dates or titles;
  • overwrite another targeted variant.

Completion checks

  • Every rendered claim resolves to one canonical record.
  • Every public claim is explicitly marked public.
  • No unresolved contradiction is presented as fact.
  • Current roles use present tense; completed roles use past tense.
  • Dates are chronologically consistent and open-ended roles are intentional.
  • The website matches the designated public variant after synchronization.
  • Generated files are reproducible from the canonical records and variant.

Skills relacionados