The Agent Skills Directory

[SAFE]: The skill focuses on documentation management (CONTEXT.md and ADRs) and does not contain any malicious instructions, obfuscated code, or unauthorized network access. It follows best practices by requiring user consent before creating files or finalizing architectural decisions.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from the conversation and the codebase, though the risk is mitigated by its restricted capabilities.
Ingestion points: Reads user conversational input (e.g., definitions of terms) and greps the codebase (via Grep and Read tools) to identify constants or model fields in SKILL.md.
Boundary markers: Employs human-in-the-loop verification; it is instructed to propose ADRs 'out loud' for user approval and confirm terms before writing to CONTEXT.md.
Capability inventory: The skill is limited to reading and writing markdown documentation files (CONTEXT.md and the docs/adr/ directory) and cannot execute code or access sensitive system configuration.
Sanitization: Includes a verification step where the agent is instructed to cross-reference user-stated behavior against the actual code using Grep to ensure documentation accuracy and consistency.

maintain-context