Skills
Navigate
skills/practicalswan/agent-skills/agent-task-mapping/Gen Agent Trust Hub

agent-task-mapping

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: No malicious code, obfuscation, or unauthorized data access patterns were detected across the skill files.
  • [COMMAND_EXECUTION]: Includes a local Node.js script (scripts/agent-discovery.js) used for identifying available agents in the workspace. The script performs local file system reads to extract metadata from .agent.md files for cataloging purposes.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the delegation of user tasks to sub-agents, creating a prompt injection surface. This is addressed by a verification protocol: 1. Ingestion points: User-provided task descriptions are ingested at runtime as per instructions in SKILL.md. 2. Boundary markers: The 'Verification Protocol' in SKILL.md explicitly mandates the use of minimal, task-local context without hidden answers. 3. Capability inventory: Sub-agent invocation via runSubagent is described in references/agent-details.md and examples/delegation-example.md. 4. Sanitization: The protocol requires manual validation of evidence, diffs, and citations before result integration.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 06:18 AM