fix: prevent duplicate append of agents_rules and cursor_rules to system_prompt #2884
Conversation
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
1A7432
force-pushed
the
fix/agents-rules-duplicate
branch
2 times, most recently
from
7779541 to
a0bc008
Compare
This commit comprehensively fixes critical bugs causing AGENTS.md and other instruction files to be duplicated in system prompts, leading to exponential token growth and context pollution. Root Causes Fixed: 1. session_ctx.system_prompt timing bug: - Previously saved at line 446 BEFORE appending agents_rules/cursor_rules - Next conversation turn would re-append these rules to incomplete prompt - Resulted in: 1x -> 2x -> 3x -> 4x... exponential growth per turn 2. Double reading when instructions_file = "AGENTS.md": - First read at line 261-270 as user message (instructions_file) - Second read at line 457 as system prompt (agents_rules) - Same file content added twice in different locations Fixes Applied: 1. Fixed session_ctx.system_prompt update timing (llm.lua:475): - Removed premature save at line 446 - Now saves AFTER all appends complete - Ensures complete prompt is cached for next turn 2. Prevented duplicate file reading (llm.lua:266-275, prompts.lua:150-175): - Track instruction_file_read to record already-read files - Pass exclude_file to get_agents_rules_prompt() - Skip files already read via instructions_file 3. Added duplicate detection as safety net (llm.lua:459-467): - Use string.find() to check existing content - Prevents append if already present - Handles edge cases and legacy sessions Impact: - Eliminates token waste and context pollution - Fixes infinite loop behavior - Maintains backward compatibility - Allows users to safely use instructions_file = "AGENTS.md" Test Scenarios: ✅ instructions_file = "AGENTS.md" - no duplication ✅ Multi-turn conversations - stable token usage ✅ Session restoration - no re-appending ✅ Legacy sessions with old prompts - graceful handling
1A7432
force-pushed
the
fix/agents-rules-duplicate
branch
from
a0bc008 to
4505a91
Compare
Root Cause: - When opts.prompt_opts.system_prompt exists (from chat_history), it was directly reused - This historical system_prompt contained old selected_files content - Rendering the template again would add current selected_files on top - Result: selected_files content accumulated exponentially per conversation turn Fix: - Always regenerate system_prompt by rendering template from scratch - Never reuse opts.prompt_opts.system_prompt from history - This ensures selected_files are only included once per turn Impact: - Fixes the critical bug where avante.md/AGENTS.md content multiplied (2x -> 3x -> 9x) - Ensures clean system_prompt generation every conversation turn - Maintains all other functionality (agents_rules, cursor_rules, custom system_prompt)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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
This PR fixes a critical bug where AGENTS.md, CLAUDE.md, and other instruction files' content would be duplicated and accumulated in the system prompt with each conversation turn.
Root Cause
In the
generate_prompts()function (lua/avante/llm.lua:457-460):opts.prompt_opts.system_promptexists, it's reused from the previous conversationagents_rulesandcursor_rulesare then blindly appended each timeChanges
agents_rulescursor_rulesstring.find()with literal matching to detect existing contentImpact
Testing
Manually tested with GLM-4.7 provider:
Related Issues
This PR addresses user-reported issues where:
References