The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic context injection pattern (!command) to execute a shell script (yaml-helper.sh) when the skill is loaded. This is used to resolve automation configurations from project YAML files.
[INDIRECT_PROMPT_INJECTION]: The skill represents an indirect injection surface because it reads and processes the content of other skills and their reference files to diagnose failures and propose fixes. If an analyzed skill contains malicious instructions designed to influence the agent's logic during the improvement loop, it could lead to unauthorized actions.
Ingestion points: Reads .claude/skills/[name]/SKILL.md and all files in its references/ folder during the Diagnosis phase (Phase 3).
Boundary markers: None identified; the skill directly processes the text of external skill files.
Capability inventory: Includes the ability to execute shell commands via Bash, write to the file system via Write, and perform file searches via Glob and Grep.
Sanitization: No specific sanitization or escaping of the ingested content is mentioned before it is processed by the model for diagnosis and fixing.
[COMMAND_EXECUTION]: The skill executes local project scripts (yaml-helper.sh) using the Bash tool to resolve configuration and potentially perform other project-specific tasks. While these scripts are part of the local environment, they represent a mechanism for shell command execution.

skill-improve