The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting data from external web searches and using it to influence agent actions and codebase modifications.
Ingestion points: Data enters the agent's context through the research:deep-researcher tool, which generates a "researcher report" from primary and recency sources on the web (SKILL.md, Step 1).
Boundary markers: The protocol does not specify the use of delimiters or sanitization steps when processing the research report during the diffing and editing stages (SKILL.md, Step 2).
Capability inventory: The agent is authorized to perform surgical edits (Edit), orchestrate other agents ("Spawn Explore agents"), and perform version control operations (commit) based on the external findings.
Sanitization: No explicit sanitization, escaping, or verification logic is provided to handle potentially malicious instructions embedded in web-based research reports.
[COMMAND_EXECUTION]: The skill provides instructions for executing local scripts and shell pipelines to maintain plugin integrity and triage stale content.
Evidence: The protocol references running a local script python scripts/lint_bundled_paths.py to enforce path naming policies (SKILL.md, Version-sensitive content).
Evidence: The "Triage on demand" section provides a shell pipeline using git log, grep, awk, and sort to generate a list of overdue plugins, writing output to /tmp/recently-touched.txt.

custom-plugin-refresh