Skip to main content

Resume Agent Overview

The Resume Agent section is the main operator handbook for this repository. Read it top to bottom if you want to configure a candidate, build the evidence pool, run the orchestrator, and verify the output without stitching the workflow together from scattered reference pages.

The actual user flow​

Most users only need to understand three actions:

  1. Run profile-bootstrap locally first to replace the repository placeholders with your actual candidate and runtime values.
  2. Only after that, add the remaining resume skills to Hermes and use pool-intake when the user wants to add experience, project, or OSS markdown files so Hermes places them in the right location and format on the VPS or wherever Hermes is running.
  3. Run resume-pipeline-orchestrator to process JDs.

After setup, users should not need to invoke jd-prefilter, jd-extraction, project-selection, point-repointing, or latex-assembly manually during normal operation. Those are orchestrator-managed stages.

Why this flow exists​

Most resume skills in this repo start as reusable templates with placeholders. That is intentional.

  • profile-bootstrap is the skill that personalizes those templates for one real candidate.
  • Until profile-bootstrap has been run and the placeholders are replaced, the rest of the skills are not candidate-ready.
  • That is why the first real setup step is local personalization, not adding every skill to Hermes immediately.
  • After the placeholders are replaced and candidate-profile is real, the rest of the skills can be added and used normally.

What the resume agent owns​

  • Candidate truth in candidate-profile
  • Pool onboarding and structure for personal projects, OSS contributions, and work experience
  • JD processing through jd-prefilter and jd-extraction
  • Evidence selection and tailoring through project-selection and point-repointing
  • Final resume generation through latex-assembly
  • Dashboard push, workflow logging, and JD queue state through resume-pipeline-orchestrator

Before you start​

Confirm all of these before trying to process job descriptions:

  • A dedicated Hermes resume profile exists for this workflow.
  • Runtime placeholders are resolved where needed by profile-bootstrap: <PROFILE_SLUG>, <POOL_DIR>, <RESUMES_DIR>, <DASHBOARD_BASE_URL>, <DASHBOARD_API_KEY_ENV>.
  • candidate-profile/SKILL.md has real candidate facts, not template placeholders.
  • Dashboard API auth is configured and the endpoint contracts in API Reference are available.
  • Evidence files exist for the candidate's work experience, personal projects, and OSS contributions.

Core operating model​

This system is not a monolithic prompt. It is a sequence of skills plus one orchestrator:

  1. profile-bootstrap helps fill candidate and runtime setup.
  2. candidate-profile becomes the candidate-specific source of truth after placeholders are replaced with real values.
  3. pool-intake and pool-versioning make sure candidate markdown files are stored in the right runtime locations and initialized in the way the pipeline expects.
  4. JD-facing skills process, select, and tailor content.
  5. resume-pipeline-orchestrator coordinates the run and handles dashboard side effects.

Read this section in order​

If you are trying to understand how to use the system as an operator, follow this order:

  1. Setup Guide
  2. Pool Content Guide
  3. Run and Verify
  4. How It Works
  5. Skill Workflow

Canonical flow​

Use this section as the complete Resume Agent handbook:

Reference pages under Setup, Pipeline, and API Reference still matter, but they are supporting material after this flow is in place.

If you want the actual skills and repository files referenced in this section, use the public source repository.

hermes-autonomous-resume on GitHub