delegate-implement

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill orchestrates subagents by aggregating user tasks, implementation scopes, and acceptance criteria into structured Markdown requests.
  • Ingestion points: Task specifications, repository files, and user objectives are ingested via standard input into scripts/run.sh and scripts/prepare.sh.
  • Boundary markers: The request structure depends on human-readable Markdown section separators (Objective, Scope, Context, Acceptance criteria, Verification, Constraints) to isolate instructions, which can be bypassed if processed code contains adversarial payloads.
  • Capability inventory: The system invokes subagent frameworks to edit local files and run arbitrary verification tests/scripts via Bash tooling.
  • Sanitization: Input escaping or filtering is absent within the provided shell shims, delegating the responsibility of input handling entirely to downstream subagent handlers and model guardrails.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 12:40 AM
Security Audit — agent-trust-hub — delegate-implement