name: dr-context-pipeline-v1 description: "Deterministic memory/context pipeline for agents: route a user message, retrieve relevant memory snippets, compress into a cited Context Pack (sources are snippet IDs), lint, and fall back safely. Prerequisite: a file-based memory layout with memory/always_on.md + topic files (works out-of-the-box with dr-memory-foundation). Use when building or standardizing agent memory, reducing prompt bloat, implementing retrieval+compression, creating a context pack, designing a memory pipeline, adding lint gates, or setting up golden regression tests for agent context."

DR Context Pipeline v1 (retrieval + compression + lint)

Use this skill to standardize how an agent loads memory into its prompt for correctness.

Prerequisites

• A file-based memory layout that includes memory/always_on.md (policy header + topic catalog) and topic files under memory/topics/.
• Recommended: install dr-memory-foundation (or implement an equivalent structure).

Operating procedure (default)

1. Load the always-on policy + topic catalog (your memory/always_on.md).
2. Route the message deterministically (task type + caps) using references/router.yml.
3. Retrieve top relevant snippets from your memory store; emit a Retrieval Bundle JSON (see schema).
4. Compress Retrieval Bundle → Context Pack JSON using references/compressor_prompt.txt.
   • IMPORTANT: Context Pack sources MUST be snippet IDs only (S1, S2, …).
5. Lint the Context Pack. If lint fails, skip compression and fall back to raw retrieved snippets.
6. Call the main reasoning model with: always-on policy header + Context Pack (+ raw snippets for high-stakes tasks) + user message.

What to read / use

• Router + caps: references/router.yml
• Compressor prompt: references/compressor_prompt.txt
• Retrieval Bundle schema: references/schemas/retrieval_bundle.schema.json
• Context Pack schema: references/schemas/context_pack.schema.json
• Golden tests starter suite: references/tests/golden.json

Notes

• Keep “always-on policy header” tiny (invariants only). Put everything else behind retrieval.
• If you need deterministic snippet IDs, follow the stable ordering guidance in references/deterministic_ids.md.