The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands including git status, git log, git diff, and rg (ripgrep) to discover repository structure and evolution. It also uses cat to write metadata to a JSON file. These operations are restricted to the local environment and align with the skill's purpose of documentation generation.
[DATA_EXFILTRATION]: The skill contains explicit security and privacy rules in references/core.md that forbid reading sensitive files such as .env, private keys, and tokens. No network operations or external data transfer mechanisms were detected.
[PROMPT_INJECTION]: The skill is subject to potential indirect prompt injection (Category 8c) because it processes untrusted content from the repository (source code, comments, and existing documentation) to synthesize new files. While references/core.md provides grounding rules and discovery discipline, it does not include explicit boundary markers or instructions to ignore embedded agent directives within the source material it reads.
Ingestion points: Reads all repository files, package configs, and git logs as specified in SKILL.md workflows.
Boundary markers: Absent; there are no specific delimiters used when the agent processes the contents of source files.
Capability inventory: Ability to write files to the openwiki/ directory and modify top-level AGENTS.md or CLAUDE.md files.
Sanitization: Instructions require grounding claims in source evidence, which provides a check against hallucination but not necessarily against adversarial instructions embedded in repository text.

wiki-init