agent preflight engine works the night shift

Make your repo easier for
AI agents to work in.

GitElf finds the things coding agents will have to guess, rediscover, or get wrong—and shows you the smallest repo change that removes that friction.

Empirically Defensible: 2026 PR study finds AGENTS.md linked to 28.6% lower median runtime & 16.6% fewer tokens (Lulla et al., arXiv:2601.20404) ↗
Try real repos: vercel/next.js expressjs/express trpc/trpc astral-sh/uv shadcn-ui/ui
--
Agent Readiness Score
Loading repo evidence…
⚠ Estimated task friction: analyzing…
5 Dimensions of Agent Readiness
Every dimension is scored directly against hard repository evidence.
Likely Agent Failure Points
Things coding agents will guess, retry repeatedly, or fail silently on:
Task Simulator
Simulate how a coding agent will traverse this repo to complete a real feature or refactor.
Presets:
Agent Preflight for this Task
--
Task: Add rate limiting to the API
Estimated Discovery Burden
Quantifiable agent friction metrics instead of vague estimates:
--
Files Traversed
--
Ambiguous Decisions
--
Likely Retries
--
Undocumented Deps
Likely Discovery Path
Files the agent must search, read, or cross-reference before writing code:
Agent Must Determine
Crucial decisions the agent must make:
Likely Friction Points
Where the agent risks getting stuck:
Triple Debt Diagnostic & Traps
Software systems exist across three interacting layers: Code (Technical), Mental Models (Cognitive), and Non-Code Artifacts (Intent). Grounded in Storey 2026 & Ahmad 2026.
⚖️
Auditing triple debt balance…
Technical Debt --
Code & Execution Layer (Cunningham / Fowler)
Package managers, tests, and CI alignment.
Intent Debt --
Goals, Boundaries & Rationale (Storey 2026)
Non-functional budgets, ADRs, executable specs, domain model.
Cognitive Debt --
Team Mental Models & Comprehension (Ahmad 2026)
Comprehension gates, verification competence, transactive memory.
Detected Intent & Cognitive Debt Traps
Where GenAI acceleration degrades maintainability, safety, and shared understanding:
Practical Remediation Toolkit
Living, low-friction repository artifacts to ground AI agents and preserve human mental models:
Agent Contract & Repair Layer
Synthesize the repo into a compact machine contract and minimal surgical diffs:
Surgical Fixes (Minimal Repo Patches)
Targeted unified diffs that remove detected friction without bloat:
Agent Mystery Test
Standardized battery testing if an agent can confidently answer 8 core questions:
Evaluating mystery test questions…
Instruction Drift Detection
Comparing existing instructions (AGENTS.md, CLAUDE.md) against live CI and manifests:
Repository Front-Door Hygiene
Standard open source contributor and documentation checklist:
The Defensible Loop: Preflight + Repair
GitElf is not merely an AGENTS.md generator—it is a continuous operating contract engine.
Repository Manifests, CI, Layout
→
GitElf Preflight Evidence & Friction
→
Agent Contract Durable Machine Asset
→
All Coding Agents Claude, Codex, Cursor
→
Drift Detection ↺ Continuous Parity