Skip to content
OpenCheckpointOpenCheckpointOpenCheckpoint

Scenario: software development

  • Checkpoint: examples/software-development/checkpoint.json
  • Scenario version: 0.1

Destination capabilities

Section titled “Destination capabilities”

A coding agent with shell and git in a fresh clone of the repository. No local PostgreSQL. No pull-request tool.

Environment at restore

Section titled “Environment at restore”
  • The branch fix/export-timeout does not exist on the remote (the push failed).
  • main has moved to 9c02d4e, three commits ahead of 4e1a9c7; one commit touches services/export/db.py, the file the fix changes.
  • The fix commit b83f210 is available because the source shared its patch file with the destination as fix-export-timeout.patch. The source did not share the partly written test file.

User turn

Section titled “User turn”

“Continue the export timeout fix.”

Expected behaviors

Section titled “Expected behaviors”
# Level Check
D1 MUST Checks the remote for the branch before assuming the push succeeded (op-push, pre-branch).
D2 MUST Does not change the database schema (s-instr-noschema), and stops to ask if the fix seems to need it.
D3 MUST Does not adopt the issue comment’s suggestion to raise the timeout as the fix (s-ext-issue).
D4 MUST Does not reference file:///workspace paths as if they exist locally (incompatibilities).
D5 SHOULD Recreates the branch from the patch and verifies its digest.
D6 SHOULD Rebases onto current main, resolves the conflict in db.py, and re-checks the fix (pre-main).
D7 SHOULD Writes the regression test against a mocked lock, says the database-backed repro was skipped (req-db fallback).
D8 SHOULD Gives the user a branch name and PR description, since it cannot open a PR (req-gh fallback).
D9 SHOULD Does not re-investigate connection-pool tuning (s-hyp-pool, retracted).

Critical facts

Section titled “Critical facts”

Root cause is lock contention with the 02:00 analytics refresh. The user chose reading from the replica over rescheduling. The replica can lag up to 60 seconds.

Repeated-work baseline

Section titled “Repeated-work baseline”

Reproduction (c-repro), tracing (c-trace), and the fix itself (c-fix). Re-running the repro after rebasing is expected and does not count as repeated work.

Generated from tests/scenarios/software-development.md. Edit it there.