The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, including output from the caveman CLI tool and local files located in .caveman/proposals/*.md.
Ingestion points: The skill reads the report_only_observations array from the caveman opportunities list command and parses local markdown files in the .caveman/proposals/ directory.
Boundary markers: The instructions explicitly warn the agent to treat local proposal files as "untrusted historic context" and to use CLI output only as diagnostic input, which mitigates the risk of the agent following instructions embedded within that data.
Capability inventory: The skill has the capability to execute the caveman CLI tool and propose/apply code changes to the repository.
Sanitization: The skill mandates verbatim use of server-provided titles and observations and requires explicit operator verification before any code changes are designed or applied.
[COMMAND_EXECUTION]: The skill executes the caveman CLI tool via the command caveman opportunities list. This is an expected functional requirement for the skill to retrieve diagnostic data. The skill specifically warns against falling back to project API keys or other authentication methods if the CLI session is not active.

caveman-optimize