recon
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub issues, pull requests, and git logs to generate implementation plans.
- Ingestion points: The skill uses
gh issue view,gh pr list, andgit log(withinreferences/issue.md,references/repo.md, andSKILL.md) to pull external content into the agent's context. - Boundary markers: The skill encourages structured output (BDD scenarios, requirements list) but does not explicitly mandate delimiters or 'ignore instructions' wrappers when reading raw issue content.
- Capability inventory: While the recon skill itself is largely read-only, it generates plans for implementation that involve file writes and potentially network operations in subsequent steps of the workflow (e.g.,
gh issue comment). - Sanitization: There is no explicit instruction to sanitize or filter the content retrieved from GitHub before processing it for planning.
- [COMMAND_EXECUTION]: The skill executes several shell commands to gather repository state and interact with GitHub.
- Evidence:
SKILL.mdusesgit status,git branch,git log,git diff, andgh pr viewto establish situational awareness. - Evidence:
references/issue.mdusesgh issue viewandgh issue commentto interact with GitHub issues. - Evidence:
references/milestone.mdusesgh apiandgh issue listto evaluate project progress. - These commands are standard development tools and are used within their intended scope for repository reconnaissance and issue management.
Audit Metadata