Source Scan

OME's core retrospective path uses native Codex and Claude session files. That is enough to create a local experience library and run high-quality evidence review without any optional source bridge.

Spool is a local AI session index that unifies Claude, Codex, Gemini, and other agent histories into one searchable pool.

Without Spool, OME still draws from native Codex/Claude sources and the current conversation; the core path is intact. With Spool, OME can add more providers as a supplemental source. Spool improves coverage for extra agents, but it is not required for a qualified Codex/Claude deep scan.

Clean Storage Model

ome source scan writes a pointer source index. It does not copy full original transcripts into OME.

The index keeps source path, provider, time, cwd, message count, and metadata hash. New scans leave summary empty by default. For older indexes, run:

ome source clean
ome source clean --yes

The first command is a dry run. The second removes legacy summaries and materialized-session markers from the source index.

Native User Evidence Index

Build a temporary user-only index before a retrospective:

ome source user-index build --provider all --json

The command prints an indexPath. Agents use that file for repeated searches:

ome source user-index search "browser validation" --index <file> --json
ome source user-index show <hit-id> --index <file> --context 4 --json

user-index is an evidence workbench for agents. It keeps user messages in a temporary index file, does not update the long-term source index, and does not summarize experience cards. The agent still owns query expansion, counterexample search, context replay, and final synthesis.

Matched Cards During Retrospectives

Prompt-time recall can still match existing OME cards while an agent is running a retrospective. That is expected: process and governance cards can help the agent scan carefully.

Those matched cards are not source evidence for a new experience. Subject-area matches should be recorded only as active-card overlap checks, such as whether the result is new, should be merged, narrowed, archived, or kept separate. The final used-card disclosure should mention only process or governance cards the agent actually applied during the retrospective.

Without Spool

Scan Codex sessions directly:

ome source scan codex --sessions ~/.codex/sessions

Agents should still expand the focus lens into several short searches before writing a retrospective audit: likely user wording, synonyms, opposite phrasing, acceptance criteria, rejection reasons, and nearby module or path names.

For qualified deep scans, prefer the native user-index path above. source scan codex remains a pointer-index command for source cataloging.

With Spool As Supplemental Source

Check availability and enable Spool as a source:

ome source status
ome source connect spool --mode enabled

Before a deep scan, check and refresh Spool:

spool status
spool sync

Prefer search-first supplemental lookup:

spool search "browser validation" --source codex --limit 10 --json
spool show <uuid> --json

Run several short searches for the same lens and record the query family in the retrospective audit. A single query such as minimal intrusion no baggage is only one probe, not evidence that the theme was fully searched. Spool hits should be treated as extra leads; core evidence still goes through native user-only indexing and original context replay when available.

Then scan only the high-value slice:

ome source scan spool --query "browser validation" --source codex --limit 50

Avoid starting a retrospective with a broad Spool scan for fuzzy topics such as architecture, fallback, or source-of-truth. Broad scans can hit very large sessions and fail after partial success. Narrow the query, inspect the result, and record any fallback in the retrospective audit.

If a high-value result is skipped because the session is large, raise the limit explicitly:

ome source scan spool --query "browser validation" --max-session-bytes 4194304

After Source Scan

Ask your agent to include scanned source records in the reflect source audit, then keep the normal draft approval flow:

Run an OME reflect from the sources we just scanned. Give me only the draft approval page, not JSON.
If I add feedback, refine the same reflect. Wait for my confirmation before adding approved experiences to the library.

Scanned material should never bypass draft approval and confirmation. Spool expands the searchable source pool; it does not change OME's safety model.