The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute specific vendor CLI commands, such as gjalla loop advance and gjalla ledger mark, to manage the review cycle and log results in SKILL.md. These commands are legitimate tools for the gjalla development environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external, untrusted content (code diffs and pull request descriptions).
Ingestion points: Code changes and metadata provided by users as referenced in SKILL.md.
Boundary markers (absent): The skill relies on persona-based instructions rather than explicit data delimiters to isolate reviewed content.
Capability inventory: The agent can execute vendor-specific CLI tools and spawn sub-agent processes as described in SKILL.md.
Sanitization (absent): No specific content filtering or escaping is mentioned in SKILL.md. This surface is inherent to the core purpose of a code review tool and is mitigated by the analytical persona defined in the skill.
[DYNAMIC_EXECUTION]: The instructions in SKILL.md mandate spawning fresh reviewers for each round to ensure objectivity. This uses standard agent sub-tasking capabilities rather than executing unverified runtime scripts.

gjalla-code-review