LIBRARY-FIRST PROTOCOL (MANDATORY)
Before writing ANY code, you MUST check:
Step 1: Library Catalog
- Location:
.claude/library/catalog.json
- If match >70%: REUSE or ADAPT
Step 2: Patterns Guide
- Location:
.claude/docs/inventories/LIBRARY-PATTERNS-GUIDE.md
- If pattern exists: FOLLOW documented approach
Step 3: Existing Projects
- Location:
D:\Projects\*
- If found: EXTRACT and adapt
Decision Matrix
| Match | Action |
|---|
| Library >90% | REUSE directly |
| Library 70-90% | ADAPT minimally |
| Pattern exists | FOLLOW pattern |
| In project | EXTRACT |
| No match | BUILD (add to library after) |
STANDARD OPERATING PROCEDURE
Purpose
- Run a disciplined, repeatable paper-reading workflow that produces searchable annotations and evidence for downstream writing.
- Apply prompt-architect refinements to surface constraints, guard against overclaiming, and keep outputs English-only with explicit ceilings.
- Preserve skill-forge structure-first requirements: SKILL.md + README + examples + references; log any gaps.
Trigger Conditions
- Positive: requests to read, digest, or annotate academic papers/books; build citation-ready notes; turn readings into evidence.
- Negative: casual skimming or non-research entertainment; route targeted prompt crafting to
prompt-architect.
Guardrails
- Extract constraints in three classes: HARD (must satisfy), SOFT (should), INFERRED (flag for confirmation) with sources cited.
- Run the two-pass refinement loop: (1) structure/coverage; (2) epistemic/confidence calibration.
- Enforce confidence ceilings: inference/report 0.70; research 0.85; observation/definition 0.95. State ceiling explicitly in every response.
- Maintain evidential trail: cite sections/pages for each insight; avoid verbatim quotes unless necessary.
Inputs
- Source details (title, authors, venue/year, access links).
- Research question or goal for the reading block.
- Time/length constraints and output format for notes.
Workflow
- Intent & Constraint Scan: Capture the research question, success criteria, and the three constraint classes; confirm INFERRED items before proceeding.
- Plan the Passes: Outline skim → detailed read → consolidation; define note schema (summary, claim, evidence, page, keyword tags).
- Pass 1 – Structural Read: Map structure, thesis, methods, key figures/tables. Log HARD coverage gaps.
- Pass 2 – Evidence Pass: Extract claims with page anchors, record confidence per claim, and tag uncertainties for follow-up.
- Synthesize & Validate: Cluster notes by theme, mark contradictions, and run a self-check against constraints and quality gates (searchability, completeness, bias).
- Package Deliverables: Produce annotated notes (Markdown), a findings digest, and open questions; store with project-level tags for recall.
Validation & Quality Gates
- Notes include source/page anchors and keyword tags; duplicates merged.
- Minimum two passes completed with documented deltas between passes.
- INFERRED constraints resolved or explicitly flagged.
- Deliverables saved alongside README, examples, and references when available.
Response Template
**Intent & Constraints**
- HARD: ...
- SOFT: ...
- INFERRED (confirm): ...
**Plan**
- Pass design and note schema.
**Findings & Evidence**
- Claim → evidence (page/link) → confidence ceiling.
- Gaps / follow-ups.
**Deliverables**
- Notes: <location>
- Digest: <summary bullets>
Confidence: 0.78 (ceiling: inference 0.70) - based on structured review of provided sources.
Confidence: 0.78 (ceiling: inference 0.70) - Template assumes structured reading completed with evidential anchors.