identity-verification

name: identity-verification description: "Verify that a named individual is who they claim to be, or resolve ambiguity between people sharing a name. Anchors on verifiable claims and cross-references independently. Writes a conforming report (per report-conventions) to /identity-verification/-.md. Requires authorisation gate." argument-hint: " claims to be <role/employer/credential> [engagement_dir]" user-invocable: true allowed-tools: WebSearch, WebFetch, Read, Write, Bash

Verify the identity of the named individual using public sources, and write a conforming report.

[!IMPORTANT] This skill requires the investigator agent's full authorisation gate before invocation. The gate stays in the session record — it is not embedded in the report file.

Step 0: Resolve arguments

Parse $ARGUMENTS as <person name> claims to be <claims> [engagement_dir]. The trailing token, if it looks like a path, is the engagement directory; otherwise default to pwd. Resolve to an absolute path.

The "claims to be" portion is the input to verify, not part of the subject identity.

Step 1: Compute the output path

Subject slug: <lastname>-<firstname>, kebab-case, lowercase, ASCII (per report-conventions).

Output path: $ENG/identity-verification/<slug>.md

If a file already exists at that path, ask the user before overwriting (overwrite / write a -2 sibling / abort).

Step 2: Stage the report from the template

mkdir -p "$ENG/identity-verification"
cp "${CLAUDE_PLUGIN_ROOT}/templates/identity-verification.md" "$ENG/identity-verification/<slug>.md"

Replace placeholders in frontmatter:

Step 3: Investigation

Write findings into the staged file as you go.

Anchor on verifiable claims

Start with what the subject has claimed — not with a general search. List each claim explicitly in the report's "Claims to verify" section before verifying any of them. Unanchored searches produce noise; anchored searches produce evidence.

Claims to verify might include:

• Current employer and role
• Credentials and qualifications
• Publications or public work
• Location or operational base

Verify each claim independently

For each claim:

Employer verification:

• Search the company's own website for the person (team pages, leadership, press releases, bylines)
• Check LinkedIn for consistency between the person's profile and the company's employee list
• Look for the person mentioned in company announcements or press coverage

Credential verification:

• Professional licensing boards (see people-lookup skill for the full AU/NZ/US registry list)
• Academic institutions for degree claims — faculty pages, alumni directories
• For specific accreditations: the issuing body's public registry or verification page
• AU health practitioners: AHPRA
• AU financial advisers: ASIC Financial Advisers Register

Publication verification:

• Google Scholar, ORCID, ResearchGate
• The specific journal or publisher's website for the claimed work
• Conference proceedings for claimed conference presentations

Cross-reference identifiers

Strong identity confirmation comes from consistent identifiers across independent sources:

• Photo consistency — does the same face appear on LinkedIn, company site, conference appearances?
• Location consistency — do professional history claims place the person in the same locations consistently?
• Timeline consistency — do claimed tenures line up? Are there unexplained gaps?
• Writing style or patterns — for authors, does the claimed work match a consistent voice or research area?

Name disambiguation

If the name belongs to multiple people:

1. List all distinct individuals found with this name
2. Apply context anchors from the gate record (employer, location, field) to narrow
3. Document which individual is being confirmed — and note if disambiguation is uncertain

Failure condition: after 3 attempts with different context anchors, ambiguity persists. Stop and ask for additional context rather than guessing.

Flag inconsistencies

Document any:

• Claims that couldn't be verified (not the same as disproved)
• Gaps in the professional timeline
• Credential claims that don't appear in the issuing body's registry
• Conflicting information across sources (different employers listed on different sites)

Distinguish clearly between "unverifiable" (no public evidence either way) and "contradicted" (evidence that the claim is false).

Raise human-required steps explicitly — but stay inside the gate

Some verification needs a paid service or human action — paid background-check providers, certified document verification, in-person interview, court-record certified copies. Silently skipping these when they're in scope is the bug to avoid. If the step is in scope per the gate record, raise it under ## Pending follow-up → Human-required steps with source, access path, rough cost, and what it would resolve.

The gate scope matters here more than in other investigator skills. This skill's job is to verify what the subject claims — not to expand into out-of-scope profiling. A paid lookup that goes beyond the gate's stated scope is not a legitimate pending item. Don't fill buckets for the sake of it. If every in-scope lookup is automated, write "None.".

If another investigator skill would resolve a gap (/investigator:people-lookup for professional context, /investigator:public-records for formal records, /investigator:social-media-footprint for photo/timeline consistency), name it under ## Pending follow-up → Skill-required steps with a one-line rationale — staying inside the gate scope.

If a source failed mid-investigation (timeout, rate limit, transient error), log it under ## Pending follow-up → Re-fetches with the URL and what it was meant to confirm.

Step 4: Finalise the report

• Every template section heading must remain in the report, even if empty. Use "Unverifiable" or "Not applicable to gate scope" rather than removing sections.
• The three sub-sections under ## Pending follow-up (Human-required steps, Skill-required steps, Re-fetches) also stay. Write "None." under any that are empty — and remember that out-of-scope lookups don't belong here.
• Set status: Final once content is complete; leave Draft if claims remain unverified or disambiguation is incomplete.
• Optionally set confidence: 0-4 based on the verification evidence (per source-quality).
• Source list with tier tags and access dates.

Rules

• Start from the subject's claims, not from an open search. You're verifying, not profiling.
• Never assert a claim is false unless you have a source that actively contradicts it. "Not found in registry" is not the same as "invalid credential."
• Disambiguation failures must be surfaced. Reporting on the wrong person is worse than reporting nothing.
• Photo comparison is visual only — note photo consistency across sources; don't speculate beyond what you can observe.
• One file per invocation.
• Silently skipping an in-scope verification step is a bug. If a paid or human-required check is in scope per the gate, raise it under ## Pending follow-up → Human-required steps. Out-of-scope lookups do NOT belong there — the section isn't a wish-list.

Output

A single line: the absolute path to the written report.