The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from public chat reports to generate GitHub issue content, creating a potential surface for indirect prompt injection.
Ingestion points: Public chat reports or summaries provided by users.
Boundary markers: The instructions mention "source boundary" and "scope boundary" as required inputs, though no specific technical delimiters (e.g., XML tags or code blocks) are explicitly mandated for the ingestion phase.
Capability inventory: The skill uses the gh CLI for searching issues and PRs and checking repository security settings via API. It also uses the omh CLI for logging status.
Sanitization: The workflow incorporates significant mitigations, including a mandatory "direction check," a limit of three follow-up questions to gather evidence, and a requirement for explicit human confirmation before any issue creation or external handoff.
[COMMAND_EXECUTION]: The skill utilizes standard command-line tools to perform its tasks.
Evidence: Use of gh issue list, gh pr list, gh api, and omh runtime record. These commands are scoped to read-only investigative actions (searching for duplicates, verifying security reporting paths) or internal platform logging.

omh-github-issue-intake