The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates the execution of a shell script located at ./.forge/skills/create-plan/validate-plan.sh. This script is executed whenever a plan is created to ensure it meets structural requirements. Executing scripts from a skill package without verification is a significant security risk as the script's contents are not audited in this context.
[REMOTE_CODE_EXECUTION]: By requiring the agent to run a locally stored script (validate-plan.sh) provided within the skill's own file structure, the skill facilitates the execution of potentially untrusted code on the host machine.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8c). It uses search, sem_search, and read tools to ingest content from the user's codebase, which is then used to generate implementation plans. If the codebase contains malicious instructions, the agent might follow them during plan generation or the validation phase.
Ingestion points: Project files read via search and read tools (documented in the 'Initial Assessment' section).
Boundary markers: None specified in the instructions to separate code content from instructions.
Capability inventory: Shell execution (validate-plan.sh) and file writing (plans/ directory).
Sanitization: No validation or sanitization of the input data or the generated file path used in the shell command is described.

create-plan