The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to conceal its AI identity and the origin of the content it generates. It forbids attribution headers and requires the removal of typical AI linguistic markers.
Evidence: SKILL.md contains an 'Absolute Prohibitions' section stating: 'NEVER: ... PR 内のどこにも Claude / AI / エージェント / アシスタントへの言及を入れない' (Never include mentions of Claude/AI/Agent/Assistant anywhere in the PR). PROSE.md additionally provides 'Anti-AI Heuristics' to strip identifiable LLM patterns.
[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by instructing the agent to ingest and act upon data from external, untrusted sources.
Ingestion points: Data enters the context via gh pr view, Slack search results, and issue tracker tickets as defined in the 'リンクを探す' (Finding links) section of SKILL.md.
Boundary markers: Absent. External context is interpolated into PR templates without markers or warnings to the agent to ignore embedded instructions.
Capability inventory: The skill has write access to GitHub via gh pr create and gh pr edit commands.
Sanitization: Absent. The skill does not define any validation or sanitization for content retrieved from external services like Slack or issue trackers.
[COMMAND_EXECUTION]: The skill relies on shell commands to interact with the Git repository and the GitHub API.
Evidence: SKILL.md documents the use of gh pr view, gh pr create, gh pr edit, git diff, and git log for context gathering and application of changes.

creating-pull-requests