The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on several local bash scripts (scripts/workspace, scripts/task-brief, scripts/review-package) and direct git commands to perform its core functions. These operations include initializing directories, extracting task descriptions from markdown files, and generating comprehensive diff packages for review. These actions are within the expected behavior for a developer-oriented autopilot tool.
[DYNAMIC_EXECUTION]: The skill defines a multi-stage process that dynamically dispatches and coordinates subagents (implementers, reviewers, fixers, and verifiers). This involves runtime assembly of prompts based on task state and file contents to guide subsequent model actions.
[INDIRECT_PROMPT_INJECTION]: The workflow ingests external data from user-supplied tickets and project specifications, which are then included in the context for subagents that perform implementation and verification tasks.
Ingestion points: User-provided specifications and ticket references are read into the plan in Phase 1 (SKILL.md).
Boundary markers: The prompt templates for subagents (such as implementer-prompt.md and delivery-reviewer-prompt.md) use brackets for interpolation but lack explicit delimiters or instructions to ignore embedded commands within the ingested text.
Capability inventory: Subagents are directed to run project test suites and execute implementation changes, creating a platform where injected instructions could potentially cause the agent to run malicious test code.
Sanitization: The skill does not perform any sanitization or validation of the input data before it is interpolated into the prompts for sub-tasks.

tracer-autopilot