scriptoriumadmin/claude-skills
1
0
Fork 0
mirror of https://github.com/alirezarezvani/claude-skills.git synced 2026-10-11 03:37:58 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
1506 commits 6 branches 7 tags 44 MiB
Commit graph

2 commits

Author SHA1 Message Date
Claude
a088c8ba77
fix: phantom-path sweep — 888 unresolvable references to zero 
A directory reorg added a skills/ path segment; hundreds of references never
followed. This sweep repoints every path-like reference in SKILL.md, agents,
commands, orchestration, and templates to verified on-disk targets:

- 30 root commands + 19 root agents: missing skills/ segment inserted
- 7 c-level persona agents: 17 hallucinated reference filenames substituted
  with the real files (e.g. okr_execution.md -> process_frameworks.md)
- 5 research skills: phantom scripts/office/validate.py step replaced with a
  runnable stdlib zip-integrity check
- email agents: skills frontmatter corrected to productivity/email
- orchestration/ORCHESTRATION.md + templates: stale paths fixed;
  agent-template now requires trigger phrasing in descriptions (root cause)
- 76 more files across engineering, c-level-advisor, compliance-os,
  research-ops, ra-qm, marketing, productivity; dual-publish pairs mirrored
- dead refs dropped/replaced where no target ever existed (REGISTRY.md,
  trend_analyzer.py, cursor-microinteractions.md)

New: scripts/check_paths.py linter (CI gate G1) + narrow allowlist for
teaching examples. Verified: 540 files scanned, 0 unresolvable.

https://claude.ai/code/session_019AJddAL1NADWMXsy1qNPQF
 2026-06-10 14:33:00 +00:00
Claude
50b5b1b9ed
feat(research): patent + syllabus — Path-B batch 3 (specialty research-pack variants) 
Slice 5 batch 3 — final two research-pack siblings. Specialty variants:
  - patent: 5-sub-use-case routing (novelty/FTO/landscape/diligence/litigation)
  - syllabus: BUNDLED-JS-DOCX-GENERATOR pattern (first in repo)

After this merges: ALL 6 research-pack siblings shipped (pulse + litreview
+ grants + dossier + patent + syllabus). 9 of 13 v2 megaprompts complete.

SOURCE SPECS

- megaprompts/11-patent-megaprompt.md (PR #657)
- megaprompts/10-syllabus-megaprompt.md (PR #657)

PATENT (Prior-Art + Landscape Intelligence)

Refuses generic "patent help". Q2 forces commitment to ONE of 5 sub-use-cases:

  novelty       → narrow + claim-text focused; verdict NOVEL/POTENTIALLY/NOT NOVEL
  FTO           → active patents only, jurisdiction-filtered; CLEAR/FLAGGED/HIGH RISK per jurisdiction
  landscape     → CPC trends + filer tally; CONCENTRATED/COMPETITIVE/EMERGING
  diligence     → assignee + assignment chain + family resolution; PORTFOLIO VERIFIED/PARTIAL/RISK
  litigation    → adjacent art before priority date; KNOCK-OUT/STRONG/WEAK/NO MATERIAL ART

Each sub-use-case uses fundamentally different search strategy (enforced by
sub_use_case_router.py). DOCX section emphasis varies per sub-use-case.

Key Path-B preserved elements:
- 6-Q grill-me intake with Q2 mandatory commitment + Q3-Q6 conditional skips
- 4 sources: Google Patents (workhorse) + Espacenet + USPTO + Lens.org BYOK
- CPC/IPC class follow-up after initial keyword search (catches keyword-missed art)
- Family resolution across jurisdictions (deduplicates same-invention filings)
- Date discipline (filing/priority/publication/grant — surface legally-relevant)
- Mandatory legal disclaimer for novelty + FTO (Q6 triggers)
- Out-of-scope flagging (trademark/copyright/trade-secret)
- 8-section DOCX with sub-use-case-specific emphasis

Scripts:
- citation_tracker.py: multi-source three-count (Google Patents + Espacenet +
  USPTO + Lens.org) + 1s sequential discipline + Lens BYOK tracking
- family_resolver.py: 3-pass clustering (family_id → priority_number →
  heuristic with 80% Jaccard on assignee + inventor + matching priority_date)
- sub_use_case_router.py: deterministic strategy from 5 sub-use-cases →
  query plan + ranking heuristic + DOCX emphasis flags + legal disclaimer flag

References (7+ sources each):
- sub_use_case_routing.md: MPEP, 35 USC 102/103, WIPO PCT, EPO Guidelines,
  USPTO PPS docs, Google Patents docs, Lens.org API
- cpc_classification_canon.md: CPC scheme, WIPO IPC, Mowery/Nelson/Sampat,
  WIPO PATENTSCOPE, Cohen/Nelson/Walsh, MPEP §901, Lemley/Sampat
- legal_disclaimer_discipline.md: MPEP §1.4-§1.5, AIPLA Code of Ethics,
  35 USC §282/§271, EPO Guidelines, PCT Article 39, Fischer/Henkel on PAEs

SYLLABUS (Course Supplementary Reading List)

Bundled-JS variant — generates .docx via scripts/generate_reading_list.js
(Node.js + docx package, ~395 lines) rather than inlining 300+ lines of
DOCX layout in SKILL.md.

Key Path-B preserved elements:
- 3-Q grill-me intake (input format + audience + year range)
- Group-and-confirm checkpoint after Phase 2 (proceed/merge/split/add/remove)
- Applied-domain weaving (e.g., "enzyme kinetics food processing" not just
  "enzyme kinetics" — boosts relevance dramatically)
- Audience calibration (undergrad-intro defines every term; grad-doctoral
  assumes technical fluency)
- Bloom higher-order discussion questions (apply/analyze/evaluate, NOT recall)
- Sequential Consensus 1 q/sec
- Source discipline + three-count tracking
- Bundled JS for DOCX (token-efficient + reusable + maintainable)

Scripts:
- citation_tracker.py: Consensus three-count + per-section breakdown +
  1s sequential discipline
- topic_grouper.py: greedy clustering of extracted topics into 6-12 sections
  via shared-keyword detection (≥2 significant words shared → same section);
  auto-merge smallest if >12, auto-split largest if <6
- discussion_question_validator.py: Bloom-level classification per question;
  flags BELOW-audience FAIL with verb-replacement suggestions; flags
  ABOVE-audience WARN
- generate_reading_list.js: BUNDLED Node.js DOCX generator (~395 lines).
  Multi-location require fallback for `docx` package. JSON input → .docx
  output. Title page + intro + learning outcomes box + numbered papers
  per section + audit log + footer.

References (7+ sources each):
- applied_domain_weaving.md: Bloom 1956, Mayer multimedia learning, Fink
  significant learning, Donald disciplinary thinking, Lave/Wenger situated
  learning, Chickering/Gamson 7 principles, Boyer scholarship of application
- audience_calibration.md: Bloom/Anderson-Krathwohl revised taxonomy, Marzano
  new taxonomy, Hattie visible learning, Bain great teachers, Walvoord/Anderson
  effective grading, Brookfield/Preskill discussion, Bjork desirable difficulty
- bundled_script_pattern.md: Karpathy-coder discipline, CLAUDE.md anti-patterns,
  docx Node.js package, CommonJS module resolution, Twelve-Factor App,
  Kernighan/Plauger Software Tools, McIlroy/Unix philosophy

REPO STRUCTURE

Both plugins in research/. Patent uses standard 11-file layout. Syllabus
uses 12-file layout (extra file: scripts/generate_reading_list.js bundled JS).

VERIFIED CLEAN

All 7 scripts pass smoke tests:

Patent:
- sub_use_case_router: FTO with US+EP → 8 queries with jurisdiction
  scaling; novelty (no jurisdictions) → 6 queries with claim-focused
  ranking
- family_resolver: 6 sample hits → correctly resolves to 3 unique families
  (Acme: 3 jurisdictions; Beta: 2 jurisdictions; Gamma: 1). Deduplication
  savings: 3
- citation_tracker: lifecycle works, multi-source counts (Google Patents
  + Espacenet + USPTO + Lens) tracked separately, audit block matches
  DOCX Section 8 format

Syllabus:
- topic_grouper: 19 sample topics → 12 sections, headings derived from
  shared keywords ("Plant + Physiology", "Animal + Anatomy", etc.)
- discussion_question_validator: 5 sample questions correctly classified.
  "What did authors find?" → recall, OK for undergrad_intro, FAIL for
  grad_doctoral with verb-replacement suggestions. "Design a follow-up
  study..." → create, OK for grad_doctoral.
- citation_tracker: lifecycle works with per-section breakdown
- generate_reading_list.js: syntax valid (node --check passes)

All scripts: --output json valid. plugin.json validates.

VERTICAL-SLICE STATUS

  ✓ Slice 1: capture (PR #659)
  ✓ Slice 2: pulse (PR #660)
  ✓ Slice 3: email pair (PR #661)
  ✓ Slice 4: landing (PR #662)
  ✓ Slice 5 batch 1: litreview (PR #663)
  ✓ Slice 5 batch 2: grants + dossier (PR #664)
  ✓ Slice 5 batch 3: patent + syllabus (this PR)
  ☐ Slice 6: notebooklm (browser-automation, last shape)
  ☐ Slice 7: 13-research orchestrator + autoresearch-agent reconciliation
  ☐ Slice 8: 02-reflect (productivity)
  ☐ Cleanup PR: move engineering/pulse + engineering/capture

9 of 13 skills shipped after this merge. ALL research-pack siblings
complete (6 of 6 in research/). Only 3 special-shape skills remain:
notebooklm (browser-auto), 13-research (orchestrator), 02-reflect
(productivity).

NOT DONE IN THIS PR (intentional)

- .claude-plugin/marketplace.json: separate concern
- .codex/skills/ symlinks: auto-sync on merge
- engineering/pulse + engineering/capture: cleanup PR queued

https://claude.ai/code/session_01FEUmeuYhmnxVFq7EZM8ZSw
 2026-05-16 02:06:56 +00:00