prior-art-search
Skill1 starsChanged 2 months agoMITView on GitHub
Mandatory pre-implementation research skill. Search GitHub, official documentation, package registries, and existing battle-tested implementations before writing new features, rules files, skills, or architecture. Enforces the Adopt-Extend-Compose-Build hierarchy and delegates heavy search loops to subagents to prevent main-thread context bloat. Use before any non-trivial feature, skill, config, or library choice — and whenever an existing approach fails or survives only on a workaround and 'keep the status quo' is tempting — not for bug fixes, refactors, or mechanical changes.
8 other projects use a version of this.
Adopted by
Most recently taken up by
- maskshell/solidforgetheir own version, first seen Nov 4, 2026
- revfactory/patent-harnesstheir own version, first seen Nov 1, 2026
- RobThePCGuy/Claude-Patent-Creatortheir own version, first seen Oct 13, 2026
- maskshell/solidforge-pitheir own version, first seen Oct 8, 2026
- talhaMah56/claude_skillstheir own version, first seen Sep 12, 2026
- wanshuiyin/Auto-claude-code-research-in-sleeptheir own version, first seen Jul 15, 2026
Used with
Projects with it also use deepxiv, invention-structuring and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
raja21068Skilldeepxiv
Search and progressively read open-access academic papers through DeepXiv. Use when the user wants layered paper access, section-level reading, trending papers, or DeepXiv-backed literature retrieval.
Also in 4 projects that use this
raja21068Skillinvention-structuring
Structure a raw invention idea into a formal invention disclosure. Use when user says \"构建发明\", \"structure invention\", \"发明构建\", \"invention disclosure\", or wants to formalize a rough idea into a patent-ready structure.
Also in 3 projects that use this
Yusong-EnceladusSkilldse-loop
Autonomous design space exploration loop for computer architecture and EDA. Runs a program, analyzes results, tunes parameters, and iterates until objective is met or timeout. Use when user says \"DSE\", \"design space exploration\", \"sweep parameters\", \"optimize\", \"find best config\", or wants iterative parameter tuning.
Also in 3 projects that use this
- +9 more from raja21068
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- wanshuiyin/Auto-claude-code-research-in-sleep★ 19ksame name
- RobThePCGuy/Claude-Patent-Creator★ 211same name
- Chanw-research/claude-code-paper-writing★ 42same name
- raja21068/AutoResearch★ 2same name
- revfactory/patent-harness★ 2same name
- maskshell/solidforge★ 1same name
- maskshell/solidforge-pi★ 1same name
- talhaMah56/claude_skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
- delegate-or-dieSkill
- doubt-driven-developmentSkill
- j-spaceSkill
- verification-before-completionSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. 0 are common, 13 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
13
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Before designing architecture, writing non-trivial features, creating skills, or synthesizing configurations (rules files, CI/CD, Docker), …
- Distinctive: line 21
If an official template, high-starred repository, or standard library solves the problem, use it directly.
- Distinctive: line 22
Do not re-implement existing standards.
- Distinctive: line 24
If a proven solution covers ~80% of the requirements, adapt only the remaining 20% for local constraints.
- Distinctive: line 27
If 2–3 small, maintained packages together cover the requirements better than one heavy dependency, combine them.
- Distinctive: line 29
Only implement original logic from scratch when research proves no satisfactory prior art exists, or the problem is genuinely novel.
- Distinctive: line 41
Query craft: search in the language of the index you are querying — for international technical content that usually means English …
- Distinctive: line 45
If a proven approach or verdict was previously recorded, reuse it immediately — sub-second retrieval, zero network calls.
- Distinctive: line 48
Check installed skills and active MCP servers.
- Distinctive: line 50
Check the official ecosystem registries (npm, PyPI, crates.io, Go modules) and official vendor documentation.
- Distinctive: line 52
Search for popular, maintained implementations, star-tested patterns, and official recipes.
- Distinctive: line 53
Search for known pitfalls, edge cases, and post-mortems for this specific domain.
- Distinctive: line 73
Every prior-art investigation must conclude with this concise structured format (<250 words):
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Prior Art Search (Search Before Building)
The best code and configurations are the ones already battle-tested by the global community.
Before designing architecture, writing non-trivial features, creating skills, or synthesizing configurations (rules files, CI/CD, Docker), search for existing solutions first. Do not invent from imagination.
Settling counts as building. When the current approach fails, breaks something else, or survives only through manual workarounds, run this skill before recommending the user keep the status quo. "No better method exists" is a conclusion only a completed search can earn — most such conclusions come from incomplete searches. Present "do nothing" alongside the realistic alternatives and justify it with the same evidence as any other verdict.
1. The Decision Hierarchy: Adopt → Extend → Compose → Build
Assess all candidates through this strict precedence ladder:
- Adopt (Direct Reuse)
- If an official template, high-starred repository, or standard library solves the problem, use it directly.
- Do not re-implement existing standards.
- Extend (Adapt & Specialize)
- If a proven solution covers ~80% of the requirements, adapt only the remaining 20% for local constraints.
- Credit the reference pattern and document the adaptation delta.
- Compose (Combine Small Pieces)
- If 2–3 small, maintained packages together cover the requirements better than one heavy dependency, combine them.
- Build (Novel Implementation)
- Only implement original logic from scratch when research proves no satisfactory prior art exists, or the problem is genuinely novel.
2. Search Priority Ladder
Execute queries in this order, stopping early when a definitive solution is found:
Before any tool call: state in plain text what functionality is needed, in which language/framework, and the constraints — so the direction can be corrected early.
Do not run this skill for: bug fixes, refactoring, config-value edits, or purely mechanical changes. If the user says research is unnecessary, skipping is itself a decision — record a one-line verdict noting it was skipped at the user's request.
Query craft: search in the language of the index you are querying — for international technical content that usually means English keywords, whatever the working language is. Never translate proper nouns, product names, or technical terms: query cloudflare as cloudflare, not as its translation.
- Persistent memory & past decisions
- Query your long-term memory tool or check
docs/decisions/(ADRs) first. - If a proven approach or verdict was previously recorded, reuse it immediately — sub-second retrieval, zero network calls.
- Query your long-term memory tool or check
- Local & installed
- Grep the current repository for existing helper functions, patterns, or utilities.
- Check installed skills and active MCP servers.
- Package registries & official standards
- Check the official ecosystem registries (npm, PyPI, crates.io, Go modules) and official vendor documentation.
- Open-source prior art (GitHub / Web)
- Search for popular, maintained implementations, star-tested patterns, and official recipes.
- Search for known pitfalls, edge cases, and post-mortems for this specific domain.
3. Subagent Execution & Context Isolation
Never pollute the main thread with raw search queries, page dumps, or scraping results.
Quick vs Full: an obvious single lookup runs inline (Quick); anything broader delegates the sweep to a subagent (Full).
- Delegate the sweep: spawn a disposable research subagent with the specific inquiry:
- "Investigate existing open-source solutions, GitHub repositories, and best practices for [TASK/DOMAIN]. Identify the top 2 battle-tested approaches."
- Disposable context: the subagent executes all multi-query searches and page fetches in its own isolated context.
- Punch-list return: the subagent synthesizes findings and returns only the structured verdict below to the main conversation.
- Status-first return: the report MUST open with one status line before any content:
STATUS: DONE|STATUS: DONE_WITH_CONCERNS(name the doubts) |STATUS: BLOCKED(name the blocker) |STATUS: NEEDS_CONTEXT(name what is missing). ABLOCKED/NEEDS_CONTEXTsubagent returns early with just the status and a one-line reason — it must not pad the report to fill the format.
4. Required Output: The Verdict Report
Every prior-art investigation must conclude with this concise structured format (<250 words):
### Prior Art Verdict
- **Status**: [DONE | DONE_WITH_CONCERNS | BLOCKED | NEEDS_CONTEXT] — first line of the report; if not DONE, follow with doubts/blocker/missing info only
- **Verdict**: [Adopt | Extend | Compose | Build]
- **Selected Reference**: [Repository / Package / Template Name] (URL or package identifier)
- **Quality Signal**: [Stars / Maintenance status / Official endorsement]
- **Adoption Strategy**:
- If **Adopt**: direct usage instructions / dependencies to install.
- If **Extend**: what 80% is reused, and what exact 20% must be written locally.
- If **Build**: explicit proof of why existing solutions are unsuitable.
- **Known Pitfalls**: [Key gotchas or anti-patterns discovered during research]One-line variant for small passes: Verdict: <Adopt|Extend|Compose|Build> — <target or "custom"> — <evidence-based reason>. A pass that searches but records no verdict line is incomplete. When the search ends in Settle (keep the status quo), the verdict needs the same rigor as a Build: name the realistic alternatives considered — including "do nothing" — and the impact of not acting.
5. Memory Write-Back (The Flywheel)
Whenever a strategic verdict is confirmed and adopted:
- Persist the core decision to your long-term memory tool, or record a local ADR in
docs/decisions/. - This ensures future sessions in this codebase build on past discoveries rather than repeating identical external searches.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

