The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests task titles and bodies from the agkan CLI to guide the implementation and self-review processes. If this task data is provided by an untrusted actor, it may contain malicious instructions intended to influence the agent's behavior.
Ingestion points: Task data is retrieved in Step 2 (agkan task get) and used in Step 4 (Implementation), Step 8 (Checkboxes), and Step 9 (Self-Review).
Boundary markers: The skill uses triple-quote delimiters when passing the task body to a sub-agent for review in Step 9, which helps delineate the untrusted data from the system instructions.
Capability inventory: The agent has access to local and remote repository operations via git and gh, and can modify task metadata using agkan.
Sanitization: The skill does not perform explicit sanitization of the ingested task content before using it in prompts or commands.- [COMMAND_EXECUTION]: The skill executes several command-line tools including git, gh, agkan, and jq. These are used appropriately to manage branches, commits, pull requests, and task status updates within the developer's environment.- [DYNAMIC_EXECUTION]: Step 9 involves a delegation pattern where a sub-agent is invoked to perform a code review. This process dynamically incorporates the task description and implementation diffs into a review prompt for a second model context.

agkan-subtask