fix: make Copilot capture attribution consistent#16
Open
codeprakhar25 wants to merge 2 commits into
Open
Conversation
The VS Code extension was using Date.now() as a session ID (so every
capture looked like a different session) and the heuristic threshold
fired on any large edit — including edits from Claude Code, Cursor, and
human copy-paste — producing silent false positives in session.jsonl.
Changes:
- extension.js: stable WINDOW_SESSION_ID per activation; confidence and
capture_mode fields on every event ("high"/manual vs "low"/heuristic);
doc comment explaining which capture modes are reliable vs. unsupported
- capture-copilot.py: passes confidence + capture_mode through to
session.jsonl; documents mode reliability in module docstring
- prepare-ledger.py: reads copilot events (kept excluded from
attribution) and builds a copilot_context summary (event count,
low_confidence flag, files, lines) for the pending payload
- finalize-ledger.py: propagates copilot_context into trace metadata
- data.rs: adds copilot_context field to AgentdiffMetadata
- list.rs: shows "~cpl" in TRUST column when low-confidence events exist
- report.rs: emits a warning note in Review Context markdown section
when low-confidence heuristic capture is detected
- test_capture_copilot.py (new): 10 tests covering all capture modes,
tool mappings, path resolution, and silent-exit behaviour
- test_extension.js: 3 new tests for confidence fields and session ID
stability
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Greptile SummaryThis PR fixes three bugs that made Copilot capture attribution unreliable: a per-call
Confidence Score: 4/5Safe to merge; one test has a logic gap that could cause a false failure in environments where AGENTDIFF_SESSION_LOG is set. The new scripts/tests/test_capture_copilot.py — unused Important Files Changed
Reviews (2): Last reviewed commit: "fix: scope Copilot warning to current in..." | Re-trigger Greptile |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| // These limitations exist because VS Code does not expose a stable public API | ||
| // that identifies the source of a document edit as Copilot vs. human vs. other | ||
| // agent. The VS Code team is tracking this at: | ||
| // https://github.com/microsoft/vscode/issues/XXXXX (placeholder) |
There was a problem hiding this comment.
The URL https://github.com/microsoft/vscode/issues/XXXXX is a placeholder and should be replaced with the real tracking issue number before this ships, or removed if no canonical issue exists.
has_cpl_warning was iterating over the full traces slice, causing the warning to appear in every intent group when any single trace had low-confidence Copilot events. Filter to only traces belonging to the current group via group.trace_ids. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
AgentDiff ReportSummary
Review Context
Files To Review First
Trace details
|
Comment on lines
+222
to
+231
| "event": event_name, | ||
| "cwd": str(repo), | ||
| "file_path": str(edited), | ||
| "model": "copilot", | ||
| "session_id": "s", | ||
| "lines": [1], | ||
| "confidence": "high" if event_name == "manual" else "low", | ||
| "capture_mode": event_name if event_name == "manual" else "inline_heuristic", | ||
| } | ||
|
|
There was a problem hiding this comment.
env dict — test is not actually isolated from environment
The env dict built on line 229 is never used; self.mod.get_session_log(str(repo)) reads os.environ directly. If AGENTDIFF_SESSION_LOG is set to a non-empty value in the process environment (e.g. by a CI runner), get_session_log follows the override path and returns a non-None path, causing the assertIsNone assertion to fail. Compare test_get_session_log_returns_path_when_initialized, which explicitly pops the variable before calling the function. The same guard is needed here.
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
Fixes #12. Copilot capture attribution was inconsistent due to three separate bugs:
vscode-${Date.now()}was evaluated on everycaptureFile()call, so each event appeared to come from a different session, breaking per-session grouping.MIN_COPILOT_CHANGE_LEN = 50fires on any large document change — including edits from Claude Code, Cursor, Codex, or human copy-paste. These false positives silently polluted session.jsonl with non-Copilot events.agentdiff.captureNowcommand) from heuristic guesses.Changes
extension.jsWINDOW_SESSION_IDper activation;confidence("high"/"low") andcapture_modefields on every event; doc comment listing reliable vs. unsupported modescapture-copilot.pyconfidence+capture_modethrough to session.jsonl; module docstring documents mode reliabilityprepare-ledger.pycopilot_contextsummary in the pending payloadfinalize-ledger.pycopilot_contextinto trace metadatasrc/data.rscopilot_context: Option<serde_json::Value>toAgentdiffMetadatasrc/commands/list.rs~cplin the TRUST column when low-confidence events existsrc/commands/report.rsscripts/tests/test_capture_copilot.py(new)scripts/tests/test_extension.jsTest plan
python3 -m pytest scripts/tests/ -v— 30/30 passnode --test scripts/tests/test_extension.js— 14/14 passcargo test— 35/35 pass🤖 Generated with Claude Code