The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external GitHub repositories, including commit messages, issue bodies, and pull request descriptions.
Ingestion points: Data is fetched via git clone, GitHub REST API, and the Wayback Machine.
Boundary markers: The skill includes 'Anti-Hallucination Guardrails' (Rule 7) instructing the agent not to run code found inside the investigated repository.
Capability inventory: The skill uses terminal commands (git, curl), file operations, and delegated sub-agents.
Sanitization: The skill instructs redacting secrets (Rule 8) but relies on agent reasoning to identify and handle potentially malicious natural language instructions within the data.
[COMMAND_EXECUTION]: The skill utilizes several terminal commands and local scripts to perform forensic tasks.
Executes scripts/evidence-store.py to manage investigation data.
Uses git commands (clone, log, fsck, reflog) for history analysis.
Uses curl to interact with external APIs.
Uses bq (BigQuery) for searching GitHub Archive data.
[EXTERNAL_DOWNLOADS]: The skill fetches forensic evidence from well-known and trusted external services.
Downloads repository metadata and events from the GitHub REST API (api.github.com).
Retrieves archived versions of web pages from the Wayback Machine (web.archive.org).
Queries the GitHub Archive dataset via Google BigQuery.

oss-forensics