The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill suite defines protocols where the agent ingests and follows instructions from local repository files, creating an indirect prompt injection surface. \n
Ingestion points: The agent is instructed to scan and read rules from lessons/INDEX.md (lessons.md), research/INDEX.md (research.md), and verify states in acceptance.md (acceptance.md). \n
Boundary markers: The specifications do not define explicit delimiters or 'ignore embedded instructions' markers for this external content. \n
Capability inventory: The instructions describe capabilities for GitHub CLI operations (gh issue/pr in board-github.md), file system management (git mv, rm -rf in conventions.md), and arbitrary shell execution for 'automatic' acceptance criteria verification (ac-spec.md). \n
Sanitization: No sanitization or validation logic is specified for content extracted from these files before processing. \n- [COMMAND_EXECUTION]: The specifications direct the agent to perform environment interactions via shell commands. \n
Evidence: board-github.md details the usage of gh issue create and gh pr create with parameters derived from document content. ac-spec.md defines 'automatic' verification criteria where the agent executes commands to observe behavior. conventions.md describes cleaning temporary work areas using rm -rf tmp/eo/.

eo-shared