v12-v13: session-mined repos + adaptive hint format#68
Merged
Conversation
HAPPY_PATHS_HINT_FORMAT=terse strips the explanation paragraph and emits only: 💡 Try: <fix command> Hypothesis: less text for the model to parse = faster action.
New repos mined from 300 real Pi sessions (~2275 errors): - monobuild: format-before-lint (533x) + build target syntax (368x) - toolhub: hallucinated tools (92x) + missing modules (88x) v13 adaptive: terse for specific commands (pip install X), verbose for discovery tasks (find setup scripts). Now 10 repos, 40 tasks, 19 unique traps.
…tasks 13 benchmark iterations, ~600+ runs total. v11 remains best general policy (-9% ledgerkit, -20% logparse median). Session mining: 14 sad path families from 300 real sessions. 2 new repos from top unmapped sad paths (monobuild, toolhub).
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Session mining + v12/v13 hint formats + 2 new benchmark repos.
What's new
Session mining — Analyzed 300 real Pi sessions (~95K tool calls, ~2,275 categorized errors) to identify the top 14 recurring agent sad paths. Top patterns: format-before-lint (533x), build target syntax (368x), hallucinated tool names (92x).
2 new repos from real sad paths:
monobuild— format-before-lint trap + build target syntax traptoolhub— hallucinated tool name trap + missing module workaround3 hint format variants tested:
Testing