The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted task data such as objectives and context via stdin, which is then processed by external AI agents. This could allow for indirect prompt injection if the input contains malicious instructions.\n- Ingestion points: Untrusted data is ingested via stdin to scripts/run.sh and scripts/prepare.sh.\n- Boundary markers: The skill uses markdown headers (Objective, Scope, Context, etc.) to structure the input.\n- Capability inventory: The skill possesses the capability to execute shell scripts and Node.js commands, which are used to process and dispatch tasks to other agents.\n- Sanitization: No explicit sanitization or filtering of the ingested content is visible in the provided shell script shims.\n- [COMMAND_EXECUTION]: The skill invokes Node.js to run a bundled delegate-cli.mjs file through shell shims. While the bash tool is scoped to these scripts in the skill directory, this involves executing internal logic that is not exposed in the markdown documentation.\n- [DYNAMIC_EXECUTION]: The shell scripts dynamically resolve the path to the Node.js entry point (delegate-cli.mjs) using dirname $0. While this supports portability and is the primary mechanism of the skill, it qualifies as dynamic loading of executable content from computed paths.

delegate-chore