-
-
Notifications
You must be signed in to change notification settings - Fork 0
Skill Catalog
Jeremy Longworth edited this page Sep 12, 2026
·
2 revisions
AgentInvestigate currently validates 212 atomic skill packages. Each package lives at:
skills/<family>/<skill>/SKILL.md
| Branch | Families |
|---|---|
| Shared professional foundation | Professional Core & Ethics |
| Private investigation | Intake, Scope & Authority; Law, Licensing, Privacy & Compliance; Investigation Planning & Case Management; Research, OSINT & Public Records; Identity, Entity & Timeline Analysis; Interviewing, Witnesses & Statements; Evidence & Chain of Custody; Investigative Analysis; Observation & Surveillance Governance; Reporting, Findings & Case Presentation; Corporate & Workplace Investigations; Background Screening & Due Diligence |
| Private security | Security Operations, Access & Patrol; Incident Response; Communication & De-escalation; Physical Security & Risk Assessment; Security Systems & Technology; Loss Prevention & Asset Protection; Investigation & Security Program Management |
The branches share selected foundations, but they are not interchangeable.
| Need | Example |
|---|---|
| Route a request |
classify-request-type, identify-jurisdiction
|
| Check authority and purpose |
assess-lawful-purpose, validate-investigative-authority
|
| Research public sources |
plan-open-source-research, record-source-provenance
|
| Organize evidence |
create-evidence-log, build-evidence-matrix
|
| Support interviews and findings |
prepare-neutral-question-set, prepare-findings-summary
|
| Support security operations |
triage-security-incident, document-alarm-response
|
| Review controls |
assess-camera-coverage-gap, identify-control-gaps
|
| Escalate regulated concerns | prepare-compliance-escalation |
- Start with routing and prohibited-capability checks when the request is new or ambiguous.
- Prefer the narrowest skill that matches the requested output.
- Read the skill's
SKILL.md, dependencies, references, and test scenarios. - Confirm the required jurisdiction, authority, privacy, source, and human-review gates.
- Preserve the skill's limitations in the resulting output.
The canonical taxonomy is docs/architecture/taxonomy-index.yaml. Run:
python scripts\validate-taxonomy.py --repo-root .
python scripts\validate-skills.py --repo-root .Do not create broad omnibus skills. A new package must be lawful, bounded, testable, non-prohibited, and absent from the existing taxonomy.