The Agent Skills Directory

[PROMPT_INJECTION]: The instructions utilize highly assertive language to define the agent's persona ("one and only deliverable", "do not skip", "non-negotiables"). It explicitly commands the agent to execute the generated plan without waiting for further approval once the initial plan is shown, which increases the potential impact of adversarial instructions found in requirements.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection as it processes untrusted external data.
Ingestion points: The agent is instructed to read Product Requirement Documents (PRDs) end-to-end to extract acceptance criteria into an EXECUTION_LEDGER.md file.
Boundary markers: There are no defined delimiters or instructions to treat PRD content as data rather than instructions, nor are there warnings to ignore embedded commands within those documents.
Capability inventory: The skill can orchestrate sub-agents, perform filesystem writes, execute Git commands (commit/push), and create Pull Requests.
Sanitization: No sanitization or validation logic is specified for the content extracted from the PRDs.
[COMMAND_EXECUTION]: The skill automates a sequence of Git operations including committing changes, pushing branches, and opening pull requests. While standard for development workflows, this execution path is triggered autonomously based on the completion of the 'Gauntlet' ledger.

the-gauntlet-glove