The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local prerequisite script at .specify/scripts/bash/check-prerequisites.sh during the initialization phase.\n- [DYNAMIC_EXECUTION]: The skill implements an extension hook system based on the .specify/extensions.yml file. It identifies and executes commands associated with hooks.before_converge and hooks.after_converge. Mandatory hooks are executed automatically using the EXECUTE_COMMAND directive, allowing external project configuration to dictate agent actions.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from the project codebase to determine its next actions.\n
Ingestion points: Reads spec.md, plan.md, tasks.md, and various source files identified during the 'Build the Intent Inventory' step.\n
Boundary markers: None. The skill instructions explicitly direct the agent to treat these external artifacts as the 'sole source of intent'.\n
Capability inventory: Includes file append operations to tasks.md and the execution of shell commands through the hook system.\n
Sanitization: There are no documented mechanisms for sanitizing or validating the content of the ingested project files before they influence the agent's logic.

speckit-converge