The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute various shell commands to manage development lifecycles and verify code integrity.\n
Evidence includes the use of pnpm typecheck, pnpm test, and pnpm lint for verification, as well as git log for checking progress (documented in SKILL.md and references/workflow.md).\n
The agent is also instructed to use the lean-spec CLI tool for project management tasks such as creating, updating, and validating specifications.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core design of reading and following instructions from project specification files.\n
Ingestion points: The agent reads specification files (Markdown with YAML frontmatter) from the specs/ directory or project workspace to extract requirements and checklist items.\n
Boundary markers: There are no instructions provided to the agent to use delimiters or safety guards when parsing these files to distinguish between legitimate requirements and potential injection attempts.\n
Capability inventory: The agent has the authority to execute subprocess commands via pnpm, git, and the lean-spec CLI tool. Notably, the lean-spec agent run <spec> command (found in references/commands.md) implies a high degree of agency driven by spec content.\n
Sanitization: No sanitization or filtering of the natural language content within specs is mentioned; the agent is explicitly told to 'Follow spec's checklist items in order' and 'Mark complete only if implementation is verified'.

leanspec-sdd