The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Node.js script using the command node ../../../.fullstack-forge/runtime/cli/src/composition-entry.js api compose --root <repository-root> --json. This command targets a script located three levels above the skill's directory, which involves executing arbitrary code on the host environment based on instructions within a markdown file.
[INDIRECT_PROMPT_INJECTION]: The skill uses a 'managed-adapter' pattern where it directs the agent to read its actual instructions ('canonical playbook') from an external path: ../../../.fullstack-forge/skills/forge-api/SKILL.md. This shifts the agent's logic to a file outside the skill's immediate package, creating a surface for instruction injection if the external file is modified.
Ingestion points: The agent is instructed to read SKILL.md from a relative path and .forge/composition.json.
Boundary markers: Absent. There are no delimiters or instructions to ignore embedded commands in the external files.
Capability inventory: The skill invokes a shell command (node) to execute local files.
Sanitization: Absent. The skill does not perform validation or escaping of the paths or the output of the executed command.

forge-api