The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions in context/PROSE_ANTI_SLOP.md direct the agent to apply stylistic rules to potentially untrusted external content, including 'PR replies', 'product copy', and 'design docs'. This establishes a surface for indirect prompt injection where malicious instructions embedded in these external sources could influence agent behavior.
Ingestion points: Pull request replies, design documents, product copy, and handoff briefs (referenced in context/PROSE_ANTI_SLOP.md).
Boundary markers: The skill does not define explicit delimiters or 'ignore embedded instructions' warnings for external data processing.
Capability inventory: The skill includes shell command execution via subprocess.run in scripts/_skill_cli.py.
Sanitization: There is no evidence of input validation or sanitization mechanisms for the external content the agent is instructed to process.
[COMMAND_EXECUTION]: The helper script scripts/_skill_cli.py implements a run_command function that uses subprocess.run to execute shell commands. This capability is provided to the agent for development tasks such as running npm-only steps, as described in the script's documentation.

layout-morph