The Agent Skills Directory

[SAFE]: The skill proactively implements security controls by requiring a 'sanitize gate' during the initialization process. It instructs the agent to gitignore sensitive files such as .env, runtime.conf, and local caches, ensuring only .example templates are committed. Additionally, it mandates a sanitization check for internal identifiers (hosts, project names, credentials) before any data is pushed to repositories.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a framework for ingesting external documentation into a centralized knowledge base, which creates a potential surface for indirect prompt injection.
Ingestion points: External data enters the wiki via the okb_snapshot.py script, which reads artifacts from a 'Source inventory' of repositories and branches during the bootstrapping of effort-based wikis (defined in references/okb-effort-wiki.md).
Boundary markers: The ingestion process uses structured snapshots (verbatim copy + frontmatter + sha256), providing metadata about the source but lacks explicit delimiters to instruct the agent to ignore instructions embedded within the ingested content.
Capability inventory: The skill performs file system writes to persistent directories (e.g., ~/handoff/, ~/ops/) and instructs the execution of local utility scripts for snapshotting and link verification.
Sanitization: The skill mitigates risks through its mandatory 'sanitize gate' invariant, which focuses on identifying and removing sensitive credentials and identifiers from the ingested data set.

kb-bootstrap