The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, specifically intake-item/v1 and idea-brief/v1 envelopes from other skills.
Ingestion points: assets/intake_request.py and assets/idea_request.py read external files to generate markdown specification skeletons.
Boundary markers: The skill uses markdown headings and variable-length backtick fences to isolate untrusted text.
Capability inventory: assets/contract_check.py can execute commands via subprocess.run() for assertion verification. assets/spec_to_tasks.py prepares tasks containing commands for later execution.
Sanitization: The skill normalizes input and escapes hidden or control Unicode characters. assets/spec_lint.py uses a 'copy tripwire' to detect commands harvested from untrusted blocks.
[COMMAND_EXECUTION]: The skill uses subprocess.run() for internal logic such as environment discovery, git interaction, and running its own testing scripts. It includes functionality in assets/contract_check.py to re-run assertions from envelopes, executing commands when specified by the user via the --rerun flag.

spec-first-planning