Improve agents-md prompt to force doc retrieval#88997
Merged
Conversation
Collaborator
Stats from current PR✅ No significant changes detected📊 All Metrics📖 Metrics GlossaryDev Server Metrics:
Build Metrics:
Change Thresholds:
⚡ Dev Server
📦 Dev Server (Webpack) (Legacy)📦 Dev Server (Webpack)
⚡ Production Builds
📦 Production Builds (Webpack) (Legacy)📦 Production Builds (Webpack)
📦 Bundle SizesBundle Sizes⚡ TurbopackClient Main Bundles: **432 kB** → **432 kB** ✅ -50 B82 files with content-based hashes (individual files not comparable between builds) Server Middleware
Build DetailsBuild Manifests
📦 WebpackClient Main Bundles
Polyfills
Pages
Server Edge SSR
Middleware
Build DetailsBuild Manifests
Build Cache
🔄 Shared (bundler-independent)Runtimes
|
gaojude
force-pushed
the
gaojude/improve-agents-md-prompt
branch
from
1883267 to
748727c
Compare
The previous instruction "Prefer retrieval-led reasoning" was too polite and agents would ignore it, defaulting to stale pre-training knowledge. The new instruction creates doubt by stating the agent's memory is WRONG, which psychologically forces the agent to actually read the docs.
gaojude
force-pushed
the
gaojude/improve-agents-md-prompt
branch
from
748727c to
aaa4f9e
Compare
huozhi
approved these changes
Jan 25, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Updates the instruction in
agents-mdCLAUDE.md output to force LLMs to actually read the docs instead of relying on stale pre-training knowledge.Before:
After:
Why This Matters
Through systematic testing with the next-evals-oss eval suite, we discovered that the original "prefer retrieval-led reasoning" instruction was ineffective. Agents would ignore it and use outdated pre-training knowledge (e.g., creating
middleware.tsinstead of the new Next.js 16proxy.tsconvention).Testing Methodology
We created an "indirect proxy" eval that tests if agents know
proxy.tsis needed for request interception in Next.js 16+, using only the prompt "Log every request to the console" (no mention of proxy/middleware).Why It Works
The phrase "What you remember is WRONG" creates psychological doubt in the model's confidence, forcing it to actually read the docs rather than trusting its pre-training.
Key elements:
Full Suite Results