The Agent Skills Directory

[SAFE]: Extensive analysis of the skill's instructions, documentation layouts, and HTML templates revealed no malicious patterns or security vulnerabilities.
[COMMAND_EXECUTION]: The skill facilitates the opening of local HTML artifacts using standard commands like xdg-open or open. These actions are intended for local user review and do not involve remote command execution or privilege escalation.
[PROMPT_INJECTION]: The skill features a surface for indirect prompt injection as it ingests project-specific artifacts such as PRDs, TDDs, and source code. Evidence chain: (1) Ingestion points: docs/better-planning/ directory, specifically <feature>-tdd.md and <feature>-prd.md. (2) Boundary markers: Not explicitly enforced in the prompt but managed via structured markdown sections. (3) Capability inventory: File read/write, HTML artifact generation, and sub-agent spawning. (4) Sanitization: No explicit sanitization of ingested text. The risk is addressed by an 'adversarial review' protocol designed to detect and resolve contradictions or erroneous instructions in the plan.
[DATA_EXFILTRATION]: The included HTML templates are designed to be self-contained and run from the local filesystem. They utilize localStorage for temporary comment storage and provide a manual export button, ensuring that no data is exfiltrated to external domains or unauthorized services.

better-planning-plan