{ "generatedAt": "2026-08-04T07:48:18.566Z", "status": "experimental", "positioning": "Context Budget Briefs help coding-agent users choose the smallest useful context for a task.", "claimBoundary": "Directional product examples only; not market proof or guaranteed savings.", "proofThreshold": "Before automation: 3 real task trials where a brief changes context selection, removes unnecessary context, or prevents a concrete retry/failure.", "count": 5, "briefs": [ { "id": "claude-md-memory-hygiene", "title": "Decision Brief: CLAUDE.md and Memory Hygiene", "shortTitle": "CLAUDE.md memory hygiene", "href": "/api/decision-briefs/claude-md-memory-hygiene.json", "markdownPath": "docs/examples/decision-briefs/claude-md-memory-hygiene.md", "scenario": "Decide what belongs in CLAUDE.md / AGENTS.md / skills vs the immediate prompt.", "audience": "Agent operator maintaining persistent project instructions", "wedge": "Separate stable rules from task-local state; verify persistent instructions are still true.", "estimatedTimeSaved": "Reduces stale-memory failures that make every future agent run worse.", "task": "Decide what should go into CLAUDE.md, AGENTS.md, reusable skills, or persistent memory versus what belongs only in the immediate prompt for a coding-agent session.", "contextBudget": "target 2k-4k · max 6k", "nextAction": "", "confidence": "Medium", "sections": [ "task", "context_budget", "recommended_context", "context_packs", "model_and_tooling", "workflow_mode", "anti_bloat_guardrails", "risks", "provenance", "confidence", "next_action" ] }, { "id": "coding-agent-model-choice", "title": "Decision Brief: Coding-Agent Model and Tool Choice", "shortTitle": "Coding-agent model choice", "href": "/api/decision-briefs/coding-agent-model-choice.json", "markdownPath": "docs/examples/decision-briefs/coding-agent-model-choice.md", "scenario": "Choose between Claude Code, Cursor, Aider, Codex, or Gemini for a large-codebase task.", "audience": "Operator choosing model/tool fit before an agent run", "wedge": "Choose by workflow shape, edit scope, tests, autonomy, and repo access — not leaderboard rank.", "estimatedTimeSaved": "Avoids reruns caused by choosing a high-context model instead of a suitable workflow.", "task": "Choose between Claude Code, Cursor, Aider, Codex, or Gemini for a large-codebase task without treating the choice as a leaderboard ranking exercise.", "contextBudget": "target 5k-12k for the selection brief and initial agent handoff · max 20k before tool/model selection is being used as a substitute for context selection", "nextAction": "", "confidence": "Medium", "sections": [ "task", "context_budget", "recommended_context", "context_packs", "model_and_tooling", "workflow_mode", "anti_bloat_guardrails", "risks", "provenance", "confidence", "next_action" ] }, { "id": "cursor-context-bloat", "title": "Decision Brief: Cursor Context Bloat / Composer Drift", "shortTitle": "Cursor context bloat", "href": "/api/decision-briefs/cursor-context-bloat.json", "markdownPath": "docs/examples/decision-briefs/cursor-context-bloat.md", "scenario": "Make UI changes in Cursor without bloating context or touching unrelated files.", "audience": "Builder using Cursor/Composer for narrow frontend changes", "wedge": "Bounded route/component/design-token context plus diff review after each loop.", "estimatedTimeSaved": "Reduces UI-agent drift by keeping broad registries and data directories out unless needed.", "task": "Use Cursor to make UI changes without bloating context or touching unrelated files.", "contextBudget": "target 6k-10k working context for the active loop. · max 18k-24k before bloat risk becomes high.", "nextAction": "", "confidence": "Medium", "sections": [ "task", "context_budget", "recommended_context", "context_packs", "model_and_tooling", "workflow_mode", "anti_bloat_guardrails", "risks", "provenance", "confidence", "next_action" ] }, { "id": "failing-ci-agent-debug", "title": "Decision Brief: Agent Debugging a Failing CI Run", "shortTitle": "Failing CI agent debug", "href": "/api/decision-briefs/failing-ci-agent-debug.json", "markdownPath": "docs/examples/decision-briefs/failing-ci-agent-debug.md", "scenario": "Have an agent fix a failing test or CI run without reopening solved problems.", "audience": "Maintainer handing a failing branch to an agent", "wedge": "Use current logs, changed files, branch diff, and prior attempts; exclude stale historical logs.", "estimatedTimeSaved": "Prevents repeated debugging loops caused by wrong branch/base assumptions.", "task": "Have an agent fix a failing test or CI run without reopening solved problems, reverting valid work, or guessing from stale branch state.", "contextBudget": "target 4k-7k · max 10k", "nextAction": "", "confidence": "High", "sections": [ "task", "context_budget", "recommended_context", "context_packs", "model_and_tooling", "workflow_mode", "anti_bloat_guardrails", "risks", "provenance", "confidence", "next_action" ] }, { "id": "large-repo-claude-code", "title": "Decision Brief: Claude Code on a Large Existing Repo", "shortTitle": "Large-repo Claude Code", "href": "/api/decision-briefs/large-repo-claude-code.json", "markdownPath": "docs/examples/decision-briefs/large-repo-claude-code.md", "scenario": "Refactor a bug-prone TypeScript module in a 40k-line repository using Claude Code.", "audience": "Senior engineer using a coding agent on an existing repo", "wedge": "Start small: active module, direct tests, failing command, recent relevant diffs.", "estimatedTimeSaved": "Turns a 2-4 hour weekly context-selection loop into a 30-60 minute kickoff pass.", "task": "Refactor a bug-prone TypeScript module in a 40k-line repository using Claude Code, without turning the session into a whole-repo context dump.", "contextBudget": "target 8k-15k working tokens · max 25k tokens before the session should be considered at high bloat risk", "nextAction": "", "confidence": "Medium", "sections": [ "task", "context_budget", "recommended_context", "context_packs", "model_and_tooling", "workflow_mode", "anti_bloat_guardrails", "risks", "provenance", "confidence", "next_action" ] } ] }