The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines patterns for constructing shell commands that incorporate user-provided text. Specifically, commands such as ssf workflow select --reason "<user choice>", ssf handoff create --objective "<objective>", and ssf state set ... dp_0_decisions "<combined summary>" use string interpolation for user input. While the skill correctly instructs the agent to validate the change-name parameter against path traversal (rejecting segments like .. or /), it does not provide similar instructions for sanitizing or escaping the content of the other descriptive string arguments before they are passed to the shell.
[INDIRECT_PROMPT_INJECTION]: The skill serves as an ingestion point for untrusted data that persists throughout the state machine's lifecycle.
Ingestion points: User-provided intents, constraints, and completion criteria collected during DP-0 and handoff creation in SKILL.md.
Boundary markers: No explicit delimiters or boundary markers are defined for the data stored in dp_0_decisions or handoff objectives to prevent them from containing malicious instructions.
Capability inventory: The skill uses the ssf CLI to update persistent state files and create handoffs, which are later read by downstream skills to guide code generation and planning.
Sanitization: The skill includes path validation for the change-name variable but lacks specific sanitization or validation logic for the free-text fields containing user input.

workflow-start