Workflow-lens sequential batch
Commission
The operator commissioned a GBrain pilot, result inspection focused on workflow, then ten further analyses one by one if the pilot was sound. Each analysis uses a fresh Sol coordinator running the standard skill, not individual jobs driven by the parent session. Commit completed work between runs so preparation sees a clean origin. After the ten runs, commission one Sol audit invocation covering all ten together, rather than ten separate auditors. Then create a workshop to investigate reusable Python tools for searching Pi sessions, informed by actual audit work. This commission does not authorize automatic worktree disposal, source changes, method changes, bypassing workflow stops, or pushing commits.
The GBrain check recommends proceeding. Its publication was integrated after the evidence commit. The batch uses the workflow-enabled method and pi-sol/medium. Retain each exact method and source revision; later integration commits change HEAD without changing method content. Runs remain isolated, and analysts receive no prior audit findings.
Selected sequence
The parent selected three recent comparators and seven additional systems to vary the workflow sample. This is a purposive sample, not a prevalence study. The first three keep their previous source commits. Other targets use standard acquisition at the default branch tip, frozen by the run.
| Order | System | Source | Requested revision |
|---|---|---|---|
| 1 | Graphiti | https://github.com/getzep/graphiti | 964bb11d1e7020e3cc67fab244268b0736b6c219 |
| 2 | Napkin | https://github.com/Michaelliv/napkin | 7582d6a46f5a11995956e60a59c41a5b242109f1 |
| 3 | Letta Code | https://github.com/letta-ai/letta-code | 271f119dcf37d58649a18e6c733386b3c710a456 |
| 4 | Dynamic Cheatsheet | https://github.com/suzgunmirac/dynamic-cheatsheet | Default acquisition |
| 5 | Instinctual Memory | https://github.com/jasonkneen/instinctual-memory | Default acquisition |
| 6 | Exo | https://github.com/exoharness/exo | Default acquisition |
| 7 | Swamp | https://github.com/swamp-club/swamp | Default acquisition |
| 8 | Fractal | https://github.com/Trampoline-AI/fractal | Default acquisition |
| 9 | Basic Memory | https://github.com/basicmachines-co/basic-memory | Default acquisition |
| 10 | DSPy | https://github.com/stanfordnlp/dspy | Default acquisition |
Acceptance and audit boundaries
Use standard run checks and stop conditions. Independently check publication integrity and the workflow's bounded account before integration; a completed engine status alone is not that check. Do not require runtime experiments or exhaustive inventories where the method permits bounded implementation findings. If a consequential blocker occurs, preserve evidence and stop that execution; no automatic rewriting of frozen outputs or state. Record the disposition before continuing or replacing a target; do not silently count a failed run as complete.
The single batch auditor should inspect accepted results and available traces, prioritize workflow usefulness and recurring operational failures, and separate source changes, model variability, repair work and measured timing. Retain local and cross-run findings separately, including successful safeguards and audit limits. Nested coordinator traces are embedded in this parent session's tool results; absence of a standalone child session file does not mean missing work. Do not implement session tools during the audit merely to complete a proposed feature list. Record recurring extraction/search needs for the later workshop.
Results
The operator brought the audit forward after five follow-up analyses on 2026-10-12: publish/integrate Instinctual Memory, commit, then audit now. The remaining five selected targets are on hold, not completed or replaced. One Sol audit invocation covers all five completed follow-ups together. GBrain remains the separately checked pilot, not a sixth batch run.
All five workflow reports were read by the parent before integration. Frozen set validation passed without failures or warnings for each; integration independently checked publication evidence before committing and merging. These checks are not a substitute for the commissioned source-and-trace audit.
Worktrees below are under /home/zby/llm/commonplace/.commonplace/worktrees/.
Their commands are in .venv/bin; each run is under its worktree's
kb/agentic-system-analyses/state/<run-id>/. Keep all worktrees and traces.
| System | Run ID | Worktree name | Integration result commit |
|---|---|---|---|
| Graphiti | AAS-2026-10-12-graphiti-1d535ca565a4-01 |
graphiti-workflow-1d535ca565a4 |
69199c6ec09fa32b13f1bbc61d844f768cc4d7d4 |
| Napkin | AAS-2026-10-12-napkin-2b85bef38252-01 |
napkin-workflow-2b85bef38252 |
83cb44a89bfb56f17efb1a141ec77eaf2f859a2d |
| Letta Code | AAS-2026-10-12-letta-code-be76afeab9b7-01 |
letta-code-workflow-be76afeab9b7 |
bb155f68660d0c7294f410e3aea4816ae3c6e2b1 |
| Dynamic Cheatsheet | AAS-2026-10-12-dynamic-cheatsheet-ca0a0b29d273-01 |
dynamic-cheatsheet-workflow-ca0a0b29d273 |
86006bf574c274b715667386ae2e3a1793538db1 |
| Instinctual Memory | AAS-2026-10-12-instinctual-memory-af48437aa60c-01 |
instinctual-memory-workflow-af48437aa60c |
74beaf33ced506f24934e99d9eb9a5cd1c3c45c5 |
The GBrain integration result was cbf65a4a38e476d59b3f0ffefb668c2faaabc094.
Batch coordinator/worker traces are nested in this parent session:
/home/zby/.pi/agent/sessions/--home-zby-llm-commonplace--/2026-10-11T15-31-10-091Z_01a12b96-de8a-711e-8a12-57ef261a7126.jsonl.
Locate each result by its run ID and nested tool results, not by assumed fixed
line numbers. Trace coverage and stable locators belong in the audit report.