fix(codex): map last_assistant_message onto Event.TaskDescription - #2070
Open
suhaanthayyil wants to merge 12 commits into
Open
fix(codex): map last_assistant_message onto Event.TaskDescription#2070suhaanthayyil wants to merge 12 commits into
suhaanthayyil wants to merge 12 commits into
Conversation
Entire-Checkpoint: 01M0DJJ1MCBA13FMEMTF9HJQ1E
Contributor
There was a problem hiding this comment.
Pull request overview
This PR improves Codex agent lifecycle event normalization by mapping Codex’s last_assistant_message field onto agent.Event.TaskDescription, allowing downstream lifecycle/strategy code to surface a concise “what just happened” description for both turn-end and subagent-stop events.
Changes:
- Populate
TaskDescriptionfor CodexStop(TurnEnd) events fromlast_assistant_message. - Populate
TaskDescriptionfor CodexSubagentStop(SubagentEnd) events fromlast_assistant_message. - Add/extend tests to cover normal,
null, and whitespace-onlylast_assistant_messagepayloads.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| cmd/entire/cli/agent/codex/lifecycle.go | Maps last_assistant_message into Event.TaskDescription for Stop/SubagentStop, with a helper to trim the nullable field. |
| cmd/entire/cli/agent/codex/lifecycle_test.go | Adds assertions and new cases to ensure TaskDescription is set correctly and stays empty for null/blank messages. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
…age fixture Address Copilot review: - trimmedLastAssistantMessage now collapses all internal whitespace (newlines/tabs) to single spaces via strings.Fields/Join, not just TrimSpace, so TaskDescription stays single-line when it propagates into commit subjects via strategy.FormatSubagentEndMessage. - Enriched the blank-message Stop fixture with turn_id, permission_mode, and stop_hook_active to match the full stopRaw schema. - Added TestParseHookEvent_Stop_MultilineLastAssistantMessage covering internal newlines/tabs. Entire-Checkpoint: 01M0DMRNA1N6982A0M76BNFD1B
Contributor
Author
|
Fixed Copilot review comments: SHA: acece6b |
… boundary Codex forwards the whole last_assistant_message into TaskDescription and SaveTaskStep puts its first 60 runes straight into the task checkpoint commit subject, so a reply opening with a credential copied out of tool output became an unredacted Git object. The same text also reached git log carrying JSON-legal NUL, ESC, C1/DEL, and Unicode bidi controls, which strings.Fields preserves and which can make rendered output disagree with what was committed. Add SanitizeSubjectContent as the single boundary for agent-supplied subject text: strip controls, format characters, and invalid UTF-8, fold whitespace runs, then redact. Redaction runs before truncation, since truncating first can cut a secret short of any rule's match and leave the prefix in the subject. Applied to subagent descriptions and TodoWrite content, so every agent is covered rather than Codex alone. Co-authored-by: Cursor <cursoragent@cursor.com> Entire-Checkpoint: 01M0KQFKAE245R4NBQWZSYCGP2
SanitizeSubjectContent now redacts before bounding input, and task descriptions are sanitized at launch, completion, and materialization so secrets do not reach task.json metadata. Co-authored-by: Cursor <cursoragent@cursor.com> Entire-Checkpoint: 01M0TTAS4JXS4JPNRZ87E9BV5N
Factory AI session titles are raw model output; sanitize before they reach UpsertCompletedTaskRecord and task.json metadata. Co-authored-by: Cursor <cursoragent@cursor.com> Entire-Checkpoint: 01M0TTKSHYGEGZK2EBBKH3SPGS
Truncate to maxSubjectRedactionInput before redact.String so whole model replies cannot force unbounded regex/entropy work on hook paths. Co-authored-by: Cursor <cursoragent@cursor.com> Entire-Checkpoint: 01M0TXKAHHBNHTN3W5734ZV6NW
Truncating before redact let secrets straddling the 4096-rune cap leak a prefix into task.json and commit subjects. Inputs over the window now return a fixed omitted label after control stripping. Co-authored-by: Cursor <cursoragent@cursor.com> Entire-Checkpoint: 01M0VA4AG4RS189AM1C71GGD9P
Check the 4096-rune cap before control stripping so padding tricks cannot shrink past the window, and omit instead of truncating redacted output. Co-authored-by: Cursor <cursoragent@cursor.com> Entire-Checkpoint: 01M0WB2VJAT5WTEM7RYPX4Q0V2
Co-authored-by: Cursor <cursoragent@cursor.com> Entire-Checkpoint: 01M0X8JH1FS6GTFQE6QM91TG7V
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
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.
https://entire.io/gh/entireio/cli/trails/1101
Scope (updated)
This branch establishes a shared cross-agent sanitization boundary before text reaches Git or checkpoint metadata:
Verification: go test ./cmd/entire/cli/strategy/... plus lifecycle/checkpoint tests for TaskDescription persistence.