agent-dev-workflow
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on executing local system commands to manage containers, git state, and process lifecycles.
- Evidence: The scripts in
references/bin/utilizedocker(to manage Postgres containers),git(for worktree and branch management),bun(for running the app and tests),asdf(for tool version management),gh(for PR status checks), and network diagnostic tools likelsoforssfor port picking. - [EXTERNAL_DOWNLOADS]: The skill fetches external tools and data required for the development environment.
- Evidence:
bin/setupinvokesasdf installto ensure required language/tool versions are present andbun installto download Node.js dependencies. - Evidence:
bin/setupandreferences/snapshots.mddescribe fetching database snapshots from Google Cloud Storage (gs://URLs) using thegcloudutility. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that could potentially contain untrusted content.
- Evidence:
bin/gc(Garbage Collector) ingests data fromgit worktree listandgh pr listto determine which worktrees are safe to delete. - Evidence:
bin/setupandbin/dbprocess database snapshots and SQL queries. - Sanitization: The scripts implement several safety gates, such as verifying git OIDs (Object IDs) before deleting branches and requiring explicit flags (
--force) for destructive actions on the canonical database.
Audit Metadata