The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing the archdev CLI tool with subcommands such as start, signals, apply, and suggest to manage collaboration sessions and modify documents. These tools are associated with the skill's author context.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted natural language data from an external Artifact channel, which presents an injection risk.
Ingestion points: External data enters the context through ask.request and agent.request signal payloads received via archdev plan collab signals (SKILL.md).
Boundary markers: No specific delimiters or instructions to ignore embedded commands are documented for the signal payloads.
Capability inventory: The agent can read sensitive plan content and perform document modifications via apply and suggest subcommands.
Sanitization: There is no evidence of input validation or sanitization for the text or intent fields provided by external collaborators before they are processed by the agent.

plan-collab