The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a project-local PowerShell script located at .specify/scripts/powershell/setup-plan.ps1 during the setup phase. The script's output is parsed to set environment variables, giving a local script control over the agent's execution context.
[DYNAMIC_EXECUTION]: The skill dynamically constructs and executes agent commands based on entries in the .specify/extensions.yml configuration file. It maps hook names provided in the YAML to specific command invocations (e.g., speckit.git.commit is transformed to /speckit-git-commit) and uses the EXECUTE_COMMAND directive to trigger them. This behavior allows project-level configuration to dictate which capabilities the agent invokes.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant indirect prompt injection surface because it relies on data from project files to determine its execution path and command logic. 1. Ingestion points: The skill reads and parses .specify/extensions.yml to identify pre- and post-execution hooks and FEATURE_SPEC for design artifact generation. 2. Boundary markers: There are no explicit instructions or delimiters provided to ensure the agent ignores malicious instructions embedded within the YAML or Markdown files being processed. 3. Capability inventory: The skill possesses the ability to execute shell commands, invoke other skills via the EXECUTE_COMMAND directive, and perform read/write operations on the filesystem. 4. Sanitization: While the skill instructions specify basic shell argument escaping for single quotes in certain scenarios, there is no comprehensive sanitization or validation for the command names, extensions, or prompts retrieved from the external configuration file.

speckit-plan