The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a protocol for processing external data that is later used in automated commands and issue creation.
Ingestion points: CLI arguments --title, --context, --labels, and --session-id (parsed from stdin) described in SKILL.md.
Boundary markers: The specification lacks definitions for delimiters or warnings to prevent the agent from following instructions embedded in the provided data.
Capability inventory: The contract describes operations including reading files via --artifact-path and interacting with GitHub through the gh CLI.
Sanitization: The contract does not specify sanitization or validation rules for the input fields, representing a vulnerability surface for the downstream implementation.
[COMMAND_EXECUTION]: The skill documents and requires the use of system commands to perform its functions.
Evidence: References execution of python3 scripts/deferred_capture.py, jq -r '.session_id', and gh issue list. These tools are used for data parsing, issue management, and verifying contract compliance.

deferred-capture