The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes git rev-parse --show-toplevel via a subprocess to programmatically identify the root directory of the current repository, enabling project-specific session filtering.
[INDIRECT_PROMPT_INJECTION]: The skill ingests historical session data, including user messages and assistant outputs, which are then interpolated into the agent's context. This represents an indirect prompt injection surface if the retrieved history contains malicious or adversarial instructions.
Ingestion points: The skill fetches transcript text from a session management server via the /api/sessions/transcript endpoint through the bundled sessions-axi script.
Boundary markers: The retrieved transcript content is structured using prefixes such as [U] for user messages, [A] for assistant messages, and [T] for tool calls, which provides basic delimitation but does not neutralize potential injection content.
Capability inventory: The skill can perform network requests to an external or local server and execute shell commands via git to gather project metadata.
Sanitization: The code does not demonstrate specific sanitization or filtering of the retrieved session history to remove instructions before the data is presented to the agent.

assist-sessions