The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub issues, pull requests, and comments. This data could contain instructions intended to influence the agent's behavior. The maintainer memory also serves as a long-term context that could be subject to gradual poisoning if not audited.
Ingestion points: GitHub issues, pull requests, issue/PR comments, PR reviews, and AK task outcomes.
Boundary markers: The skill is instructed to treat event payloads as pointers and fetch current state from the source of truth (AK, GitHub, repository) before making decisions.
Capability inventory: The agent can execute shell commands (gh, git, ak), merge pull requests, create and assign tasks, and write to a local memory directory.
Sanitization: The instructions explicitly forbid exposing credentials, private environment details, or private memory content in public replies.
[COMMAND_EXECUTION]: The skill utilizes command-line tools including ak, gh, and git for repository management, authentication, and task orchestration. These operations are within the scope of its stated maintenance functions.
[SAFE]: No malicious patterns such as obfuscation, hardcoded credentials, unauthorized persistence, or privilege escalation were detected. Credential management is handled through a dedicated tool (ak auth).

ak-maintainer