本文へ移動
cccskills
無料GitHub で公開

source-driven-development

Ground framework-specific implementation in official documentation — detect stack versions, fetch authoritative docs, implement cited patterns, cite sources. Load when building with a library or framework where API correctness matters, when the user wants verified or documented code, or before writing framework code from memory. Also triggers on "source driven development", "cite the docs", "check official documentation", "verify against docs", "don't hallucinate APIs". Not for pure logic or renames. Pairs with feature-spec and test-driven-development.

インストール方法を見る

含まれるファイル(3)

  • SKILL.md5.6 KB
  • references/examples.md2.0 KB
  • references/source-hierarchy.md1.1 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Source-Driven Development

You implement framework-specific code from official documentation for the detected version, not from training-data memory. Every non-trivial framework decision is cited or flagged unverified.

Hard Rules

Read dependency manifests (package.json, pyproject.toml, go.mod, etc.) before writing framework-specific code. Fetch the specific doc page for the feature — not the framework homepage. Never cite Stack Overflow, random blogs, or training data as primary authority. If versions are ambiguous, ask once — do not guess. Surface conflicts between docs and existing codebase; do not silently pick one. Mark patterns you could not verify as UNVERIFIED explicitly.


Workflow

Step 1 — Detect stack and versions

Read the project's dependency file. State findings explicitly:

STACK DETECTED:
- [package] [version] (from [file])
→ Fetching official docs for [feature].

If versions are missing, ask the user before implementing.

Step 2 — Fetch official documentation

Fetch the relevant documentation page for the exact API or pattern. Use references/source-hierarchy.md. Prefer hooks/sdd-cache (Claude Code) or python3 .agents/skills/research-skill/scripts/doc_cache.py "<url>" — see research-skill → references/doc-cache.md.

Extract: API signatures, recommended patterns, deprecations, migration notes. If official sources conflict, surface the discrepancy to the user.

Step 3 — Implement documented patterns

  • Use signatures and patterns from the fetched docs.
  • Prefer current recommended APIs over deprecated ones from memory.
  • When docs conflict with existing project code, present options A/B and wait for preference.

Step 4 — Cite sources

In code (framework-specific lines only):

// useActionState for form pending state — React 19
// Source: https://react.dev/reference/react/useActionState#usage

In chat: full URLs, quote non-obvious passages, note browser/runtime support when relevant.

If no official doc exists for a pattern:

UNVERIFIED: No official documentation found. Based on training data — verify before production.

When NOT to use

  • Renames, typos, file moves with no API surface change
  • Pure language logic identical across versions
  • User explicitly wants speed over verification ("just do it quickly")

Gotchas

  • Training data lags; "confident" APIs are often deprecated.
  • Fetching the whole docs site wastes tokens — one page per decision.
  • Simple snippets become copy-paste templates — wrong patterns spread fast.
  • Version skew: React 18 patterns in a React 19 repo break silently.

Common Rationalizations

ExcuseReality
"I'm confident about this API"Confidence is not evidence. Verify signatures against current docs.
"Fetching docs wastes tokens"One wrong API costs hours of debug time.
"Docs won't have what I need"Absence means the pattern may not be officially recommended — flag it.
"I'll add a disclaimer instead"Either cite or mark UNVERIFIED — hedging helps nobody.
"Simple task, no need to check"Simple wrong patterns get copied across the codebase.

Output Format

## Source-driven — [feature/slug]

Stack: [detected versions]
Docs fetched: [URLs]
Implementation: [summary]
Citations: [list or UNVERIFIED items]
Conflicts with codebase: [none | options A/B]

Examples

<examples> <example> <input>Add React 19 form submission with pending state.</input> <output> Detect React 19.x from package.json. Fetch react.dev/useActionState. Implement with cited comment. Reject manual useState+isPending if docs recommend useActionState unless user chooses consistency with existing code. </output> </example> </examples>

Verification

  • Versions read from dependency file (or user confirmed)
  • Official docs fetched for each framework-specific pattern used
  • Code matches current-version documented APIs (no deprecated APIs without flag)
  • Non-trivial decisions have full URL citations or UNVERIFIED labels
  • Doc vs codebase conflicts surfaced to user

Red Flags

  • API choice made from training data not fetched docs
  • Entire docs site fetched instead of one decision page
  • Deprecated API used because snippet was memorable
  • Version in code mismatches version cited from docs

Reference Files

  • references/source-hierarchy.md: Authority order and non-authoritative sources — read at Step 2.

Prune Log

Last pruned: 2026-07-04

  • No changes — citation audit passed; content current (improve-skills full pass 2026-07-04)

Impact Report

Feature: [slug] | Stack: [versions]
Docs: [count] fetched | UNVERIFIED: [count]
Conflicts surfaced: [yes/no]

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Put on the adversarial hat and systematically attack any document, plan, strategy, or idea to expose its weakest points before commitment. Structured devil's advocate with red team rigour — not pessimism, but evidence-based critique across three phases: diagnostic (are claims accurate?), creative (is the problem artificially constrained?), challenge (are solutions robust?). Load when the user asks to stress test a document, red team this plan, poke holes in this, devil's advocate this, challenge my assumptions, or when product-soul, brainstorming, prd-writing, or inversion calls for adversarial review. Also triggers on "what am I missing", "what could kill this", "find the flaws", or "critique this rigorously".

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Design execution structure for decomposed processes: single agent or multi-agent topology. Load when user says "design an agent for this", "what agent structure do I need", "architect this", "should this be multi-agent", "what's the right execution structure", "agent topology", "how should agents be organized". Takes process-decomposer output as primary input. If triggered directly without a process entry, calls process-decomposer first.

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Internal skill. Called by setup-evaluation after a PASS. Launches agents from a validated architecture spec using Claude Code / Ampcode native parallelism (Task tool). Does NOT generate scripts or SDK code — it outputs structured spawn instructions that the platform executes natively. Never invoked directly by the user. Never launches without a setup-evaluation PASS.

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Sync library skills from an agent-loom upstream repo into this project's .agents/skills while preserving project-local and forked skills. Load when the user asks to sync agent-loom, update skills from upstream, rsync from ../agent-loom, pull new library skills, upgrade installed skills, or refresh the .agents folder without losing custom project skills. Also triggers on "sync skills from agent-loom", "update my agent skills", "pull skill library updates", or "merge agent-loom improvements into this repo".

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Instrument a shipped product's AI agents with tracing and observability so you can see what they did, why outputs happened, and what each run cost. Plain-language primer plus free-tier-first backend selection (Langfuse, Phoenix, LangSmith, Braintrust) and OpenTelemetry/OpenInference instrumentation. Load when the user asks to add observability, add tracing, instrument my agents, see what my agent is doing in production, set up Langfuse or Phoenix or LangSmith, debug why my agent gave a bad answer, or track LLM cost per request. Also fires when agent-system-architecture or setup-evaluation requires an observability plan for an agent-chain product. NOT for tracing the coding agent itself — that is run-trace. Precondition for runtime-learning-loop.

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

Run a structured retrospective after development-phase runs of your product's agents — interview the owner in plain language about what went well and poorly, draft ranked improvement hypotheses, then design and run small n=1/n=2 experiments with pre-declared success criteria, guardrails, stop conditions, and a cost/ROI kill-switch. Load when the user says how did that run go, retro this run, the agent output was bad, what should we improve, draft hypotheses, run a small experiment, or after repeated dev runs of an agentic system produce uneven quality. Priority: output quality over performance over cost, each with diminishing-returns stops. NOT a product A/B test (experimentation), NOT coding-agent harness repair (harness-evolution), NOT production-scale learning (runtime-learning-loop).

日本語の概要は準備中です。原文の説明を表示しています。

dvy1987/agent-loom32026年8月8日 更新

dvy1987 のスキルをすべて見る

このスキルの問題を報告する