lockedin-audit

lockedin-audit

Scores and refines existing resume or cover-letter documents against the same rubric used by render-resume-en and render-jaso. No vault is required for the drive-by path.

Entry points

A. Drive-by audit (no vault required)

User provides a file path. The skill:

1. Calls lockedin ingest <path> --dry-run via Bash to extract text without mutating the vault.
2. Detects document type (English resume vs. Korean cover letter (jaso)) from content and file name.
3. Routes to the correct rubric (see RUBRIC.md) and runs the selected mode (default: Score only for drive-by).
4. Returns a markdown report. No vault writes. No file writes.

B. Post-ingest audit (existing vault user)

Triggered when lockedin ingest has just completed and merged N new entities into the vault. The skill presents a 4-option menu:

Ingest complete. <N> new entities merged into vault.

What next?
  [1] Score only          — apply rubric to the source document as-is.
                            No vault changes.
  [2] Refine only          — diff-based refinement of merged entities
                            (normalize metrics, replace weak action verbs,
                            restructure highlights). Every change shown as
                            diff, user approves before vault write.
  [3] Refine then Score    — refinement followed by scoring of refined
                            output (DEFAULT RECOMMENDED — quantifies score
                            lift)
  [4] Skip

Three operating modes

Mode 1 — Score only

Applies the rubric to the source document and emits a clean markdown report. Steps:

1. Detect document type (English resume or Korean cover letter (jaso)).
2. Load the appropriate rubric via RUBRIC.md routing.
3. Writer turn (prompt-scorer.md): reads the source, quotes specific lines as evidence, produces a draft score narrative.
4. Reviewer turn (prompt-reviewer.md): runs in a SEPARATE Claude context, re-loads RUBRIC.md fresh, emits a JSON score.
5. Merge writer narrative and reviewer JSON into the final markdown report (see Output format).