The Agent Skills Directory

[DATA_EXFILTRATION]: The skill transmits project information including the project slug and the absolute filesystem path (via pwd) to an external service called "Atrium". The instructions describe using POST /api/dev-projects/sync and POST /api/dev-projects/{slug}/dep-decisions to sync this data.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources that could be controlled by an attacker, such as package manifest files and vulnerability audit reports.
Ingestion points: Project manifest files (package.json, pyproject.toml, requirements.txt, Cargo.toml, go.mod) and outputs from npm audit, pip audit, and cargo audit in SKILL.md (Step 1-3).
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potentially malicious content within these files.
Capability inventory: The skill utilizes Bash for shell command execution and curl for network requests.
Sanitization: While the skill uses jq to parse structured JSON data, it does not sanitize the values within that data before presenting them to the agent for decision-making.
[COMMAND_EXECUTION]: The skill uses Bash to execute multiple package management and auditing tools (npm, pip, uv, cargo, go, pip-audit, cargo-audit, cargo-outdated). These commands are triggered based on the presence of local files.
[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch dependency decisions and project history from the Atrium service (GET /{slug}/dep-decisions/overdue, GET /api/dev-projects/{slug}/history). The destination URL resolution logic is not explicitly defined in the provided file.

deps