The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow example script in SKILL.md contains shell injection vulnerabilities. The variables $file, $name, and $rows are used in shell contexts (e.g., wc -l < "$file" and JSON payloads) without escaping or validation, allowing an attacker to execute arbitrary commands if they can control these variable values.
[DATA_EXFILTRATION]: The skill utilizes pilotctl send-file, which can be directed to transmit any local file. This poses a risk of data exfiltration if the agent is manipulated into sending sensitive system configuration, credentials, or private keys.
[PROMPT_INJECTION]: The skill facilitates communication with external peers and processes their responses via pilotctl --json received. There is a risk of indirect prompt injection where malicious instructions embedded in dataset metadata or peer messages could influence the agent's behavior.
Ingestion points: pilotctl --json received (SKILL.md)
Boundary markers: None identified; messages are processed as raw JSON.
Capability inventory: pilotctl (file transfer, messaging), shell operations (wc, head).
Sanitization: None; the scripts perform direct string interpolation for command arguments.
[PROMPT_INJECTION]: There is a metadata discrepancy where the author is listed as vulture-labs in the frontmatter, while the README links and documentation point to TeoSlayer. This inconsistency can be deceptive regarding the skill's origin.

pilot-dataset