forge-guard

name	forge-guard
description	AI application screening -- detect AI-generated resumes, verify claims, surface authenticity signals
user_invocable	true
args	mode

forge-guard -- Router

Mode Routing

Determine the mode from {{mode}}:

Input	Mode
(empty / no args)	discovery -- Show command menu
Resume text or file path	scan
scan	scan
batch	batch
verify	verify
compare	compare
interview	interview
report	report
calibrate	calibrate
dashboard	dashboard

Auto-scan detection: If {{mode}} is not a known sub-command AND contains resume-like text (keywords: "experience", "education", "skills", "responsibilities", job titles, company names, date ranges), execute scan.

---

Discovery Mode (no arguments)

Show this menu:

forge-guard -- Application Screening

Available commands:
  /forge-guard {resume}     --> SCAN: full AI detection on a single application
  /forge-guard batch {dir}  --> Batch screen a folder of applications
  /forge-guard verify {resume} --> Cross-reference claims against public sources
  /forge-guard compare      --> Compare resume vs LinkedIn vs cover letter
  /forge-guard interview {resume} --> Generate verification interview questions
  /forge-guard report       --> Screening report for hiring committee
  /forge-guard calibrate    --> Calibrate thresholds with known samples
  /forge-guard dashboard    --> View screening stats

Or paste a resume directly to run a full scan.

---

Context Loading by Mode

After determining the mode, load:

All modes load:

• modes/_signals.md (signal taxonomy)
• config/thresholds.yml (score_ranges for PASS/REVIEW/FLAG and per-signal weights; shipped in-repo — customize via calibrate, then run npm run verify)

Then load the mode file:

• modes/{mode}.md

scan mode

Also load templates/report-template.md when saving a per-candidate screening report (modes/scan.md Step 6).

batch mode

Also load templates/report-template.md so each per-candidate report matches standalone scan output (modes/batch.md Step 4 and modes/scan.md Step 6).

report mode

Also load templates/report-template.md so committee output stays aligned with per-candidate report shape, headings, and PASS / REVIEW / FLAG semantics (see modes/report.md Step 1).

calibrate mode

Also load examples/authentic-resume.md and examples/ai-generated-resume.md when the user needs checked-in practice samples (synthetic personas, not real candidates; see modes/calibrate.md Input).

dashboard mode

Follow modes/dashboard.md Step 1: read completed rows from data/screenings.md when building aggregates. That tracker is gitignored hiring data (see Guardrails); if the file is missing or empty, use the short notice path in the mode file instead of inventing stats.

Execute the instructions from the loaded mode file.

---

Guardrails

• Screening output is probabilistic. Do not present detection as proof of AI use or dishonesty.
• Never auto-reject a candidate from score alone. Low scores mean prioritize human review.
• PASS / REVIEW / FLAG follow score_ranges in config/thresholds.yml exactly as in modes/scan.md Step 4. Qualitative authenticity bands in CLAUDE.md are narrative context only; they do not override those cutoffs.
• Every flag needs concrete evidence (quotes, counts). Do not fabricate verification results; use WebSearch or provided sources only.
• Data hygiene: Do not commit candidate PII or hiring artifacts upstream — data/screenings.md and reports/ are gitignored by design (see repo .gitignore and CLAUDE.md).
• Full workflow, scoring bands, and report placeholders: see CLAUDE.md in the repo root.

Repo changes (maintainers)

From the repo root, run npm run verify before committing upstream changes to modes, config/thresholds.yml, templates, examples, or this skill. The script checks that signals, thresholds, docs, and templates stay aligned.

For optional non-interactive passes that run verify and commit in a loop, use scripts/cursor-agent-loop.sh (see the script header and README.md Development). For verbose JSON stream output from that driver, use scripts/cursor-agent-stream-format.py (Python 3; see its header).