to-seams
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and store unvalidated text (the 'agreed interface') which is explicitly intended to be read by other agents as 'ground truth' in their prompt context.
- Ingestion points: User-provided interface text pasted verbatim (Step 3) and seam name/paths (Step 2).
- Boundary markers: Absent. The skill specifies that the interface file content goes into the consumer's prompt 'byte-for-byte' without parsing or markup.
- Capability inventory: The skill performs file writes to the local workspace (
.claude/seams/and.claude/command-centre.toml) and executes shell commands for workspace discovery. - Sanitization: Absent. Instructions mandate pasting the user's input 'verbatim' and specifically advise against summarizing it.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command to verify the workspace configuration and locate the repository root.
- Evidence:
ls "$(git rev-parse --show-toplevel)/../.claude/command-centre.toml" 2>/dev/nullin Step 1. - Context: This is used for environment discovery to ensure the necessary configuration files exist before proceeding.
Audit Metadata