The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mentions the use of standard system utilities like sha256sum (Linux) and shasum -a 256 (macOS) in SKILL.md and reference/report-template.md. These are intended for generating cryptographic hashes of evidence to ensure a chain of custody, which is a standard forensic practice and does not pose a security risk in this context.
[DATA_EXFILTRATION]: The skill provides explicit instructions to prevent data exposure. Step 7 in SKILL.md mandates the redaction of uninvolved third parties and sensitive credentials (national-ID numbers, payment card numbers, plaintext passwords) before publication, promoting secure data handling.
[INDIRECT_PROMPT_INJECTION]: As an OSINT reporting tool, the skill is designed to ingest and process untrusted external data (web pages, registry records, artifacts).
Ingestion points: Raw artifacts, HTML, and OSINT findings are ingested into the report structure as described in SKILL.md and the templates in reference/report-template.md.
Boundary markers: The skill uses a structured template with clear section headers to separate observations from assessments, which helps organize data but does not provide specific escape sequences for untrusted content.
Capability inventory: The skill metadata includes disable-model-invocation: true, which significantly reduces the attack surface by preventing the agent from executing tools even if malicious instructions were present in the ingested data.
Sanitization: The skill emphasizes the redaction of sensitive identifiers and the use of cryptographic hashes to maintain integrity without exposing raw secret data.

write-the-intel-brief