The Agent Skills Directory

[SAFE]: The skill implements a robust 'diff-before-write' discipline that requires the agent to surface proposed changes to the user before applying them. This provides transparency and a manual verification gate for all modifications.
[SAFE]: The skill requires verifiable source citations for every new claim, reinforcing data provenance and preventing the insertion of unsourced or fabricated information.
[SAFE]: Tool usage is restricted to standard file system operations (Read, Write, Edit, Bash) for local wiki maintenance. The use of the Bash tool is specifically documented for grep-based searches and does not involve risky network operations or privilege escalation.
[SAFE]: The skill manages the risk of processing untrusted data (Indirect Prompt Injection) through explicit workflow steps.
Ingestion points: Step 2 ingests data from external URLs, local files, or user-pasted text to update wiki pages.
Boundary markers: The skill uses a structured 'diff-before-write' protocol and side-by-side comparison as a conceptual boundary before execution.
Capability inventory: The skill utilizes Edit, Write, and Bash tools to modify markdown files and logs within the designated wiki directory.
Sanitization: Risks are mitigated through the human-in-the-loop requirement where the agent must describe the change and its justification before proceeding.

wiki-update