The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches methodology documentation from well-known technical sources including GitHub, Atlassian, and sites maintained by industry experts such as Martin Fowler and Sandi Metz. These fetches are restricted to a predefined list in references/external-sources.md and are used to inform planning decisions without executing remote code.
[PROMPT_INJECTION]: The skill processes untrusted data from repository files, Jira task plans, and GitHub issue descriptions, creating an attack surface for indirect prompt injection. * Ingestion points: The agent reads docs/<KEY>-tasks.md, docs/<KEY>.md, and various source code files in the repository to generate planning artifacts. * Boundary markers: The skill contains explicit instructions across all files, including SKILL.md, references/data-contracts.md, and subagent definitions, to treat external data as data only and not as instructional content that could override agent behavior. * Capability inventory: Capabilities are limited to reading and writing to the local docs/ directory and performing network fetches from a whitelisted set of URLs. The skill does not have shell execution or arbitrary system write capabilities. * Sanitization: The skill utilizes natural language delimiters and explicit instructions to the agent to ignore any embedded instructions found within the processed data.

planning-task-execution