The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting external, potentially untrusted data to guide its actions.
Ingestion points: The skill receives an "authorized Changeset Tree, acceptance criteria, constraints, and supporting solution proof" as primary inputs.
Boundary markers: The instructions do not define specific delimiters or escaping mechanisms to prevent these inputs from influencing the agent's logic beyond the intended scope.
Capability inventory: The agent has the capability to perform file system mutations ([CREATE], [MOVE], [UPDATE], [DELETE]) and execute shell commands for git management and verification.
Sanitization: No specific sanitization or validation of the changeset content is described before implementation starts.
[COMMAND_EXECUTION]: The skill requires the execution of shell commands to perform its core functions.
Git Management: Explicitly uses git reset --soft HEAD~1 for safe rollbacks.
Toolchain Invocations: Triggers typecheckers, linters, and test suites based on the configuration in .agents/environment.md.
[SAFE]: The skill documentation includes an explicit security invariant to protect sensitive environment data: "Never inspect or modify secret files (.env*, *.key, *.pem)". This proactively limits the agent's access to credentials during its operation.

forge