The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection as it ingests untrusted data from ambient conversation and existing repository documentation (CONTEXT.md, ADR files). However, the risk is significantly mitigated by the skill's narrow scope and strict procedural requirements:
Ingestion points: Reads existing CONTEXT.md and docs/adr/ files from the repository root, as well as ambient context from user interactions.
Boundary markers: Explicit boundary markers for ingested files are not defined in the instructions, but the skill requires an "offer" step to show proposed changes to the user first.
Capability inventory: The skill uses Write and Edit tools to modify text-based documentation files; it does not possess network or shell execution capabilities.
Sanitization: No specific sanitization logic is provided, relying instead on the user's manual review of the proposed changes.
[COMMAND_EXECUTION]: The skill is limited to basic file operations (Read, Write, Edit, Grep, Glob) and does not utilize shell access or command-line tools that could be abused for privilege escalation or malicious activity.
[DATA_EXFILTRATION]: There are no network tools (curl, wget) or external URL references within the skill. Data operations are restricted to the local filesystem for the purpose of maintaining project documentation.

domainkit