The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external advisory feeds which represent an attack surface for indirect prompt injection if the source were compromised.
Ingestion points: lib/feed.mjs fetches advisory data from a remote URL or reads from a local file.
Boundary markers: The data is validated against a schema and verified using Ed25519 signatures before processing.
Capability inventory: The skill has network access (fetch), file system write access within ~/.hermes, and the ability to modify user crontabs.
Sanitization: The isValidFeedPayload function performs strict schema validation, and signature verification ensures data integrity from the trusted source.
[PERSISTENCE]: The skill provides scripts to maintain recurring security checks by modifying the user's crontab.
Evidence: scripts/setup_attestation_cron.mjs and scripts/setup_advisory_check_cron.mjs use spawnSync to update the system crontab.
Context: This is a documented primary feature of the skill designed to automate security monitoring. It uses clear markers (# >>> hermes-attestation-guardian >>>) to manage its block and requires an explicit --apply flag from the operator.
[EXTERNAL_DOWNLOADS]: Downloads advisory updates from the vendor's infrastructure.
Evidence: lib/feed.mjs fetches data from https://clawsec.prompt.security/advisories/feed.json.
Context: This is expected behavior for an advisory-tracking tool. The downloads are verified using pinned public keys.
[COMMAND_EXECUTION]: Executes system commands to manage scheduling and verification.
Evidence: Uses spawnSync to call crontab and openssl for artifact verification.
Context: These executions are scoped to the skill's security functions and are triggered by user-initiated scripts.

hermes-attestation-guardian