source-driven-development
Skill1 starsChanged 2 months agoMITView on GitHub
Ground version-sensitive code in docs fetched this run, not recalled API shape — resolve the installed version, fetch the matching page, cite the URL in the code. Use it before routing, forms, data fetching, auth or a platform API. Not whole-diff review (`code-review`) or dependency CVEs (`security-and-hardening`).
71 other projects use a version of this.
Adopted by
Most recently taken up by
- ishandutta2007/Awesome-Agent-Skillstheir own version, first seen Nov 8, 2026
- TheAstroByte/astrobyte-agent-skillstheir own version, first seen Nov 8, 2026
- MMACABRE/SKILLStheir own version, first seen Nov 8, 2026
- steven123397/nexuskit-skillstheir own version, first seen Nov 1, 2026
- AgnesAI-Labs/agnet-skills-cntheir own version, first seen Nov 1, 2026
- yuusakuri/agent-skillstheir own version, first seen Nov 1, 2026
Used with
Projects with it also use constraint-driven-development, verification-before-completion and 8 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.
MrCipherSmithSkillconstraint-driven-development
Use when a project needs a persistent, measurable quality bar — starting a project without documented standards, 'set up constraints', or when coverage/perf/a11y thresholds get re-debated per-PR. Produces CONSTRAINTS.md with enforceable numbers, ratchets, and a guard against silent weakening. NOT for throwaway prototypes or a one-off CI setup.
Also in 13 projects that use this
OpenAISkillverification-before-completion
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always
Also in 6 projects that use this
ElasticSkillaccessibility
Accessibility guidance for Kibana. Use this skill when working with or reviewing EUI components, resolving a11y-related (@elastic/eui) ESLint issues, and ensuring proper use of ARIA attributes, focus management, keyboard interactions, and accessible naming conventions.
Also in 3 projects that use this
PostHogSkillkarpathy-guidelines
Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria. Based on https://github.com/multica-ai/andrej-karpathy-skills.
Also in 3 projects that use this
OpenAISkillbrainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.
Also in 5 projects that use this
CloudflareSkillreceiving-code-review
Use when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation
Also in 3 projects that use this
- +3 more from openai
Other versions71 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 71
- addyosmani/agent-skills★ 114ksame name
- shashankswe2020-ux/whoop-mcp★ 174same name
- tranhieutt/software_development_department★ 71same name
- drvoss/everything-copilot-cli★ 48same name
- vinvcn/addyosmani-agent-skills-zh★ 46same name
- jmxt3/gitscape.ai★ 33same name
- sampleXbro/agentsmesh★ 27same name
- jaskirat1616/grok-skills★ 22same name
- ishandutta2007/Awesome-Agent-Skills★ 21same name
- galando/temper★ 18same name
61 more not listed
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 repository41 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- acceptance-criteriaSkill
- aim-higherSkill
- api-designSkill
- architecture-designSkill
- browser-testing-with-devtoolsSkill
- ci-cdSkill
- codebase-designSkill
- codebase-researchSkill
- code-reviewSkill
- code-simplificationSkill
- comprehension-quizSkill
- debugging-and-error-recoverySkill
- deprecation-and-migrationSkill
- documentation-and-adrsSkill
- doubt-driven-developmentSkill
- environment-manifestSkill
- gauntlet-loopSkill
- git-workflowSkill
- handoffSkill
- idea-refineSkill
- incremental-implementationSkill
- interview-meSkill
- literate-explainerSkill
- observability-and-instrumentationSkill
- orchestratorSkill
- performance-optimizationSkill
- plan-breakdownSkill
- preflight-readinessSkill
- project-setupSkill
- pull-requestSkill
- quality-verificationSkill
- security-and-hardeningSkill
- shipping-and-launchSkill
- spec-grillingSkill
- spec-reviewSkill
- test-driven-developmentSkill
- to-prdSkill
- using-agent-skillsSkill
- worktreeSkill
The file
Files
Source-Driven Development
Purpose
Stage: Implement. Principles 6, 7, 8.
Grounds a version-sensitive decision in a page fetched this pass rather than recalled API shape: resolve
the installed version, fetch that feature's page, build against the signature it shows, cite the URL
beside the code. incremental-implementation or test-driven-development pulls it in mid-slice; the
citations ride that caller's diff as evidence a code-cold review can check without the maker's context.
Recall holds patterns that read as current and were replaced two majors ago.
When to use / when to skip
- Version-sensitive code is about to be written — routing, forms, data fetching, auth, state, config, build setup, a browser or runtime API.
- The sentence forming is "I think the signature is…".
- Framework code is being read for stale patterns, a dependency just crossed a major, or two APIs solve the same thing and the version's own docs say which one it ships.
- Skip renames, typos, file moves and pure logic — a version changes nothing there.
- Near-miss — five-axis grading of a whole diff:
code-review, which reads these citations. Dependency CVEs and supply chain:security-and-hardening. Which dependency to adopt:architecture-design.
Inputs
- the dependency manifest (
package.json·pyproject.toml·go.mod·Cargo.toml) — helps: the installed version, what makes a pattern current or stale · without it: the lockfile, the vendored package, the imports, the runtime the scripts invoke — say which,derived docs/features/<slug>/environment.mdruntime-deprows — helps: cross-checking the stack, never the version (the rows carry no values) · without it: the manifest and the imports,deriveddocs/features/<slug>/plan/<slice-id>.md— helps: which page to fetch — an API reference, not a homepage · without it: the code about to be written,derived- a way to fetch a page — helps: the document itself · without it: the vendored types and source in the
tree, every unconfirmed pattern marked
UNVERIFIED:
At most three questions with a person present, only where the version stays unresolvable and a major boundary decides the pattern.
The Process
- Resolve the version before fetching, and say where each came from — the manifest for the declared range, the lockfile for what resolved. A range is not an installation.
Fetch the page for the feature you are building, at the version you resolved — that hook's API reference, that route helper; never the homepage, never the whole site. Rank what you fetch:
Rank Source 1 official documentation 2 the project's blog or changelog 3 web standards — MDN, the spec 4 compatibility data — caniuse, node.green — never: Stack Overflow, a tutorial, an AI-written summary, recall - Read it for the deprecation and migration notes, not only the happy-path snippet — the signature, the required options, whatever it marks legacy. Carry the browser or runtime support data with any platform feature behaviour rests on.
- Build against the signature you fetched, taking the documented current way and leaving the deprecated one alone.
- Mark whatever you could not confirm
UNVERIFIED:on the line itself, naming what you looked for and did not find. - Settle a disagreement between two official sources against the version you resolved — prefer the version-specific page, or run the smaller claim once against the installed version to see which holds; record the settlement as step 7 does.
- Default to the documented pattern where the docs disagree with the repository's own code, say why,
name the codebase-consistent alternative in a Decided-for-you row, and append one
docs/session-log.mdentry. Only the stop list (safety-rails.md) ends a slice. - Cite every version-sensitive pattern where the code is, two comment lines —
// <framework> <version> <feature>then// Source: <full URL>. Full URLs, never a shortener, anchored deep where the page has one — an anchor outlives a restructure a top-level page does not. Quote the passage carrying a decision the code alone leaves unobvious.
Rationalizations
| The excuse | The reality |
| "I'm confident about this API." | Confidence tracks familiarity, not the version on disk. |
| "Fetching a page costs tokens." | An hour on a signature that moved costs more than one fetch. |
"^18 in the manifest is close enough." | A range is a permission; the lockfile says what shipped. |
| "I'll say in the message it might be stale." | A hedge in prose is gone next session; an UNVERIFIED: marker rides the diff. |
| "The repo does it the other way — I'll ask which." | The repo's way was written against some version; an unattended question stalls the slice, and a default with a reason reverses from the pull request. |
Red flags
- A framework call written before any version was resolved.
- "I believe" or "I think" standing where a URL should be.
- A homepage, a search page, a Stack Overflow answer or an AI-written summary cited for a signature.
- A deprecated API the version's own migration note names, still in the diff.
- A hedge in the prose where the line carries no
UNVERIFIED:marker. - A conflict resolved with no reason and no Decided-for-you row.
Verification
- The resolved version stands with the file it came from; anything inferred is marked
derived. - Every version-sensitive line traces to a page fetched this pass, cited beside it in the two-line shape by full URL.
- Every cited page is official documentation, a changelog, a web-standards reference or compatibility data.
- No deprecated API survives the diff, against that version's migration notes.
- Whatever the docs left uncovered carries an
UNVERIFIED:marker naming what was looked for. - Every conflict, document against document or documents against the repository, is settled with a reason and a Decided-for-you row.
Outputs & handoff
- Citations and
UNVERIFIED:markers in the caller's diff — the two comment lines of step 8, keyed off downstream by theSource:token. No artifact of its own and noSTATE.mdrow: the caller owns the slice's transition (state-schema.md). docs/session-log.md— one entry per settled conflict or derived version, cap 60 words, shape instate-schema.md. Report the measured count against the cap; trim an over-cap entry rather than hand it on, and where the file does not exist hand the entry back forproject-setup.- Returned in conversation — the resolved stack with each version's source, the URLs fetched, the
UNVERIFIED:list with its reasons, the Decided-for-you rows, and any question a person present would have been asked. - Verdict — none: this pass grades nothing.
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.

