feat: add impact analysis guidance to agent prompts #179
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
Enhance prompts to leverage the graph DB for impact analysis and change planning - a key differentiator against text-based code search tools. Changes: - codebase_understanding.j2: Add "Impact Analysis" section teaching agents to query callers and dependencies before modifying code - cypher_query_patterns.j2: Add 6 new patterns for multi-hop traversal, blast radius analysis, dead code detection, and risk assessment - plan.j2: Make impact analysis mandatory for existing codebases, add required "Impact Analysis" section format - tasks.j2: Add dependency-aware task ordering by blast radius 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <[email protected]>
Update prompts to make clear that: - Shotgun agents create plans and tasks - Coding agents (Claude Code, Cursor, etc.) execute them - Impact analysis informs better planning, not direct code changes 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <[email protected]>
Only show impact analysis guidance in prompts when a codebase graph
is actually indexed. This prevents confusing agents with instructions
for tools they can't use.
Changes:
- Add has_codebase_indexed field to AgentDeps
- Check for indexed graphs before rendering system prompt
- Wrap impact analysis sections in plan.j2, tasks.j2, and
codebase_understanding.j2 with {% if has_codebase_indexed %}
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <[email protected]>
Update test mocks to properly handle the new has_codebase_indexed field: - Add AsyncMock for codebase_service.list_graphs_for_directory - Add has_codebase_indexed attribute to mock fixtures - Update assertion checks to expect has_codebase_indexed parameter - Patch add_system_prompt_message in exception handling tests 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <[email protected]>
- Remove "check codebase before answering any question" - too aggressive - Replace verbose 6-step workflow with focused 3-step version - Emphasize "graph queries are cheap, file reads are expensive" - Add explicit "NEVER read files to understand the codebase" This should reduce context bloat from excessive file exploration. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <[email protected]>
- "NEVER read files" was too broad - Changed to "Avoid reading many files just to explore" - Allow reading files user mentions or that are identified as relevant 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <[email protected]>
Contributor
|
Hi, would it be possible to pull this into the project? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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
codebase_understanding.j2teaching agents to query callers/dependencies before modifying codecypher_query_patterns.j2for multi-hop traversal, blast radius, dead code detectionplan.j2with required section formattasks.j2(low-risk first, high-risk last)Why This Matters
This activates the graph DB as a competitive differentiator against Claude Code and other text-based tools:
Test plan
uv run ruff check .)uv run mypy src/)test_telemetry_settings_defaultsmay fail due to env-specific PostHog key)🤖 Generated with Claude Code