lsm / neokai / 26006208895 / 21
81%
dev: 81%

DEFAULT BRANCH: dev
Ran
Files 287
Run time 8s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 7.603% (+0.004%) from 7.599%
26006208895.21

push

github

web-flow 
Add persistent agent memory (#1931)

* feat: add space agent memory store

Persist explicit agent memories per Space and surface them through MCP, RPC, db-query, and task prompt injection so future sessions can reuse learned project context.

* fix: tighten agent memory search semantics

Use FTS5 ranking for memory search, preserve filtered pagination, and avoid rebuilding the memory index on every daemon startup.

* fix: reduce agent memory FTS churn

Index memory keys and restrict FTS trigger work to content-bearing updates so access telemetry stays cheap.

* fix: restore agent memory runtime tools

Attach agent memory MCP servers across restore and post-approval paths, and reject non-finite memory RPC numeric inputs.

* fix: gate memory MCP requirement

Only require the agent-memory MCP server for workflow sessions when persistent memory is configured.

* fix: cap agent memory prompt size

Limit stored memory content and injected prompt snippets so oversized agent-written memories cannot bloat workflow prompts.

* fix: preserve memory metadata and broaden FTS query

- Preserve existing tags on memory updates when caller omits tags; explicit empty array still clears.
- Keep original created_by_session immutable on conflict so provenance stays with the original author.
- Retain path separators (slash, colon) in normalized FTS query terms so paths and URLs match through the trigram tokenizer.

* fix: harden agent memory write boundaries

- MCP memory.write forwards undefined tags so omitted updates preserve stored tags instead of clearing them.
- RPC agentMemory.write does the same and no longer trusts createdBySession from request payloads, eliminating provenance forgery.
- RPC numeric params (limit, offset) must be safe integers, so values outside the SQLite range are rejected with a clear validation error instead of a SQLite datatype mismatch.

* fix: cap per-memory tag length to bound prompt size

Tags are interpolated verbatim into agent prompts ... (continued)

6363 of 83689 relevant lines covered (7.6%)

3.99 hits per line

Source Files on job daemon-online-coordinator - 26006208895.21