ss-multi-repo-workflow

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its ingestion of untrusted external content.
  • Ingestion points: The skill processes PRD links, issue tracker links, and plain-text requirements as described in the 'Inputs' and 'Input Type Detection' sections of SKILL.md.
  • Boundary markers: There are no explicit instruction delimiters or 'ignore embedded instructions' warnings present in the orchestrator to isolate user-provided requirement text from the execution logic.
  • Capability inventory: The skill manages subprocesses through 'headless process' or 'native child' adapters, performs file writes (distributing sub-plans), and conducts network operations via Git and PR-hosting CLI tools (gh, glab).
  • Sanitization: No evidence of sanitization, escaping, or validation of the external requirement content is specified before it is interpolated into sub-plans and passed to per-repo execution runners.
  • [COMMAND_EXECUTION]: The skill dynamically executes repository-scoped runners using a variety of process adapters.
  • Evidence: The 'Runner adapter' input and 'Launch a batch' process in SKILL.md define how the orchestrator launches ss-coding-workflow in separate directories.
  • Context: While the skill provides architectural guardrails—such as Iron Rules that prevent the orchestrator from editing source code directly and Pre-flight Checks to verify repository paths—it fundamentally relies on the dynamic execution of nested agent sessions in the host environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 05:56 PM
Security Audit — agent-trust-hub — ss-multi-repo-workflow