The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically identifies and executes shell commands for building and testing. It prioritizes commands found in openspec/config.yaml (test_command, build_command) and falls back to standard runners in package.json, pyproject.toml, or Makefile.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from the workspace to determine its execution flow.
Ingestion points: Reads tasks.md, design.md, specs/, and configuration files (package.json, openspec/config.yaml).
Boundary markers: Absent; instructions in these files are treated as trusted parameters for verification logic.
Capability inventory: Executes arbitrary shell commands and performs file system search/read/write operations.
Sanitization: No validation or escaping is performed on the command strings or file paths retrieved from the workspace.
[REMOTE_CODE_EXECUTION]: The skill instructions include a step to dynamically load additional skills from a path provided in the launch prompt (Step 1: Load Skills), which constitutes dynamic loading from a computed path.
[COMMAND_EXECUTION]: The skill uses mem_save to persist verification reports, which involves writing potentially untrusted execution output back to the system's memory or filesystem depending on the mode.

sdd-verify