workbench-memory-synthesis
Workbench Memory Synthesis
Use this skill for durable memory, session summaries, decision logs, issue closeout synthesis, and handoffs.
Memory Rules
- Prefer short canonical updates over duplicate drafts.
- Keep the next session resumable from files, not hidden model memory.
- Record decisions separately from logs.
- Keep strategy in synthesis documents and execution evidence in logs or issue comments.
- If a session-derived behavior should affect future agents, bridge it through
workbench-waking-upinto a public-safe repo surface instead of leaving it only in chat memory. - Preserve IDs that matter for recovery: issue IDs, run IDs, comment IDs, agent IDs, autopilot IDs, commit IDs.
- When a run stalls and a conductor proxy posts evidence, record both the proxy ID and the later agent-authored artifact ID if it appears.
- Record live resource IDs only after post-sync verification confirms them.
Synthesis Method
- Collect the latest verified state.
- Remove stale next actions that are already complete.
More from fearvox/multica-ultimate-workbench
workbench-design-docs
Product design, technical design documents, user-facing copy, specs, diagrams, and handoff documentation.
7workbench-token-context-discipline
Compact context, cache-aware execution, scoped evidence reads, and role-specific skill attachment discipline.
6workbench-gsd-tasking
Owner-scoped task decomposition with gates, rollback, verification commands, and smoke tests.
6workbench-implementation
Minimal-patch implementation, ops changes, scripts, automation specs, and local repository edits.
6workbench-research
Evidence-first research, architecture discovery, vendor documentation review, benchmark reading, and unknown-system analysis.
6workbench-code-review
Findings-first review discipline for code, diffs, task plans, live workflow changes, and implementation evidence.
6