Skip to main content

Feature Request: RW Temporary Execution Log for Recovery and Forensics

Type: Feature Request (FR)
Identifier: FR-044
Submitted: 2026-02-22
Submitted By: User
Priority: MEDIUM
Severity: MEDIUM
Status: ACCEPTED

Summary​

Add a temporary execution log file to the Release Workflow (RW) that is written during RW execution. The log enables (a) recovery of aborted runs (e.g. IDE crashes, process kills), and (b) forensic analysis of what happened during a run.

Problem Statement​

Context:

  • Some projects (e.g. Starborn Legacy) experience IDE or process crashes more frequently than others
  • When RW is aborted mid-run (crash, kill, sandbox failure, etc.), there is no persistent record of what completed
  • User must reconstruct state manually: which steps ran, what changed, where to resume

Current Situation:

  • RW runs as an agent with no durable execution log
  • Cursor chat / TODO list may be lost on crash
  • No structured artifact to support resume or forensic review

Impact:

  • Aborted runs leave partial state (e.g. version bumped but not committed)
  • Recovery is ad hoc and error-prone
  • No forensic trail for debugging workflow failures

Proposed Solution​

RW writes a temporary execution log during its run:

  1. Log file location: Configurable (e.g. {project_root}/.rw-execution.log or rw-config.yaml path); default in project root
  2. Log format: Structured (e.g. timestamp, step number, step name, status, key outputs)
  3. Log contents:
    • Each step start/end with status (PASS/FAIL/SKIP)
    • Key decisions (version bump, task attribution, paths used)
    • File changes (version file, changelog paths, etc.)
    • Errors and warnings
  4. Lifecycle: Log is appended during run; optionally rotated or truncated at start of new run; user may retain for forensics
  5. Recovery use: On resume or manual recovery, user (or agent) can read log to determine last completed step and current state

Forensic use:

  • Post-mortem on aborted runs
  • Debugging workflow behavior
  • Auditing what changed in a run

Acceptance Criteria​

  • RW creates/opens a log file at start of execution
  • Each RW step writes start and completion (or failure) to the log
  • Log includes sufficient context for recovery (step number, key outputs, file paths)
  • Log path configurable via rw-config.yaml (optional)
  • Documented in RW agent execution guide
  • Log retained for at least the duration of the run; retention policy (keep/rotate/truncate) TBD