agentic-spec-to-knowledge-derivation

name	agentic-spec-to-knowledge-derivation
description	Use after creating or materially updating a Superpowers Spec to derive durable L1/L2/L3 business assets, run deterministic lint/doctor, apply semantic doctor/repair when needed, and prepare a phase confirm request.

Agentic Spec-To-Knowledge Derivation

Required Context

Load these files before deriving, doctoring, or repairing assets:

• docs/solutions/l1/arch/business-asset-knowledge-model.md
• skills/agentic-spec-to-knowledge-derivation/rules/derivation.md
• skills/agentic-spec-to-knowledge-derivation/rules/agentic-memory-alignment.md
• skills/agentic-spec-to-knowledge-derivation/rules/secret-redaction.md

Workflow

1. Read the source Superpowers Spec and compute its repository-relative path.

2. Read docs/assets/builder/asset-index.yaml when it exists and verify the source Superpowers Spec is listed as a formal asset_kind: spec asset for the phase.

3. If the source Superpowers Spec is not listed, add a fallback derivation change that creates or registers a formal asset_kind: spec asset for the original source Spec, with repo-relative path, current source hash, phase id, and asset-index membership. This fallback records the source document itself; it must not rewrite, move, or delete the original source Spec.

4. Read relevant existing L1/L2/L3 assets under docs/solutions/.

5. Use the templates under templates/ to add, modify, or delete durable business assets.

6. Use agentic-memory family conventions for source fidelity, durable knowledge shape, and archive-compatible structure. Do not shell out to agentic-memory CLI from this skill.

7. Write L1/L2/L3 asset changes only after the user has requested derivation or repair.

8. Run deterministic CLI lint or doctor:

   agentic spec-derivation lint --manifest docs/assets/builder/derivations/<phase-id>.yaml <workspace>
   

   Module fallback:

   PYTHONPATH=python python -m agentic_core.cli spec-derivation lint --manifest docs/assets/builder/derivations/<phase-id>.yaml <workspace>
   

9. When long_document_mode triggers, create a source chunk index, dispatch read-only chunk extraction subagents when available, collect source-chunk-fact-packet outputs, synthesize the source coverage ledger as the main agent, and block if any critical chunk remains incomplete.

10. In long-document derivation, preserve source_chunk_index_ref and source_coverage_ledger_ref in the manifest and keep business-flow.operations grounded in source-backed operation contracts.

11. If deterministic lint fails for structural reasons, run CLI repair once.

12. If business meaning is missing or inconsistent, run semantic doctor and semantic repair using prompts/doctor.md and prompts/repair.md.

13. Re-run deterministic lint after semantic repair.

14. Prepare the confirm request for Agent Chat / agent-runtime.

Guardrails

• Do not modify or delete the Superpowers Spec during derivation, doctor, or repair.
• Derivation and repair must not commit, tag, push, reset, rebase, stash, or force push.
• Semantic repair may modify or physically delete only L1/L2/L3 business assets.
• 主 agent 是唯一 synthesis owner。Read-only subagents may extract chunk facts, but they must not write manifests, ledgers, specs, or final assets.
• Confirm must revalidate manifest, asset index, source hash, source refs, dirty paths, Git guard, and phase metadata before committing.