The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is explicitly designed to execute a local shell script (~/.claude/evolution/bin/evolve-promote.sh) using the bash interpreter. It does this both for checking gates and for performing the final promotion.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from candidate files, creating a potential injection surface.
Ingestion points: The skill reads candidate metadata and content from YAML files located at ~/.claude/evolution/candidates/<id>.yaml.
Boundary markers: There are no explicit instructions or delimiters defined to separate the candidate data from the agent's instructions, nor are there warnings to the agent to ignore embedded instructions within those files.
Capability inventory: The skill has the ability to execute shell scripts (bash) and display file diffs to the user.
Sanitization: The protocol does not specify any sanitization or validation of the <id> parameter or the candidate file content before it is processed or passed to the shell script.

promotion-gate