The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to interact with the caveman CLI tool through shell commands like caveman learn report and caveman mem. These commands are essential for measuring token usage and managing memory, and their execution is explicitly gated by per-edit user consent.
[PROMPT_INJECTION]: The skill processes external JSON data from CLI outputs and local files in ~/.caveman/candidates/, creating a surface for indirect prompt injection.
Ingestion points: The agent ingests token sink reports and remediation candidates from the local filesystem.
Boundary markers: The skill does not explicitly define delimiters to separate this external data from the system prompt.
Capability inventory: The agent possesses capabilities to read/write configuration files and execute shell commands via the specified CLI.
Sanitization: The skill implements a robust integrity check requiring the agent to verify the SHA256 hash of any code block against a locator before performing modifications, ensuring that file content has not been altered since the initial scan.

caveman-learn