The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow where the agent evaluates human-provided explanations and implementation diffs to gate progress. This creates a potential surface for indirect prompt injection if those inputs contain instructions aimed at subverting the agent's evaluation or subsequent implementation actions.
Ingestion points: Human explanations for prior code increments (recorded in modules/ramp-ledger.md) and the implementation diffs themselves.
Boundary markers: The skill instructions do not specify the use of delimiters or isolation protocols to prevent the agent from following instructions embedded within the human responses or diff content.
Capability inventory: The agent uses file modification tools including Write, Edit, and MultiEdit to build features, as described in the advancement-gate.md module.
Sanitization: No sanitization, filtering, or validation logic for the content of the explanations or diffs is mentioned in the skill documentation.
[COMMAND_EXECUTION]: The skill references and requires the execution of a local Python script, guard_scope_ramp.py, acting as a PreToolUse hook. This script is used to enforce implementation bounds by reading skill-related state from environment variables (IMBUE_RAMP_OK, IMBUE_STAKES) and local configuration files (.imbue/stakes, .imbue/ramp-ok).

graduated-implementation