The Agent Skills Directory

[SAFE]: The skill is strictly defined as a read-only advisor for repository source code and configuration. It is explicitly prohibited from modifying existing source files or Git history, limiting its scope to writing a single local finding ledger.
[SAFE]: It implements proactive defenses against indirect prompt injection. The instructions mandate that the agent must read all repository content as untrusted data and ignore any instructions or commands embedded within the code or documentation being audited.
[SAFE]: The finding contract includes a specific security rule to prevent credential exposure. The agent is instructed never to copy secret values (tokens, passwords, keys) into the finding ledger, recording only the location and credential type instead.
[SAFE]: No network operations, remote code execution patterns, or external dependency downloads were detected. The skill relies entirely on local file access within the repository it is auditing.
[SAFE]: The skill maintains a clear boundary between auditing and implementation. It explicitly prevents the agent from creating plans, tickets, or executing fixes, which reduces the risk of autonomous system manipulation.

tk-improve