The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard search utilities like grep, ripgrep, and ast-grep, alongside a domain-specific omh CLI for indexing. These commands are used for local repository analysis and do not present a security risk within the context of the workflow.
[CREDENTIALS_UNSAFE]: Explicit instructions are provided to redact secrets and prevent the exposure of credentials found within the repository's configuration files, ensuring a secure-by-default approach to codebase analysis.
[INDIRECT_PROMPT_INJECTION]: The skill implements a mandatory evidence chain to mitigate risks from processing potentially untrusted codebase data: it identifies the repository files as the primary ingestion point, establishes clear 'not-evidence' boundary markers to prevent hallucinated architecture claims, inventories search and indexing capabilities, and requires the redaction of sensitive values as a sanitization step.
[DATA_EXFILTRATION]: No network exfiltration patterns or unauthorized external communications were detected. Network-related activities are limited to internal runtime telemetry within the host framework.

omh-codegraph-refresh