You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 26308d1
Browse filesBrowse the repository at this point in the historyBrowse files
Copy file name to clipboardExpand all lines: CHANGELOG.md
+1Lines changed: 1 addition & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -12,6 +12,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
12
12
13
13
### Fixed
14
14
15
+
- **The post-tool-use hook no longer collapses escape-heavy tool outputs to `{}`.** The client-side 32k wire bound sliced strings by raw character count, ignoring that JSON escaping (quotes, backslashes, control characters) can double or triple the serialized form. An oversized truncated string then failed the object accounting's cost check, and the loop's `break` discarded every remaining field — outputs like `{stdout: "<quote-heavy json>", file_path: ..., exit_code: 0}` arrived as `{}`, losing exactly the small fields the structure-preserving bound exists to keep. The hook's truncation (now in `src/hooks/truncate.ts`, shared by the built copies in `dist/hooks/` and `plugin/scripts/`) mirrors the server's `boundValue`: strings are cut by binary search on serialized length, object fields are taken smallest-first so a huge `stdout` cannot crowd out `file_path`/`exit_code`, and a single unfittable entry is skipped instead of discarding the rest. Field order of over-budget objects changes in transit only; the server re-bounds and re-orders anyway.
15
16
- **Fresh installs are portable and persistent (aligned with upstream #892).** The bundled `iii-config.yaml` uses cwd-relative paths and the engine was spawned without a cwd, so on global and npx installs `./data/state_store.db` and `./data/stream_store` landed in whatever directory the user ran the CLI from, and the `iii-exec` supervision block never resolved — nothing respawned the worker after it died, which surfaced as "all data gone" reports against a live REST port. `agentmemory start` now regenerates a runtime config each boot (`~/.agentmemory/data/iii-config.runtime.yaml`) with absolute data paths and an absolute worker exec line, anchors the engine cwd at `~/.agentmemory`, and copies any legacy `./data` stores from the invocation directory on first run. Repo checkouts keep the cwd config and repo-root cwd, so dev behavior is unchanged; user overrides via env or `~/.agentmemory/iii-config.yaml` are passed through verbatim. The viewer port is now configurable (`III_VIEWER_PORT`, default REST + 2), worker pidfile paths honor `AGENTMEMORY_RUNTIME_DIR`, Docker installs gain uid/gid/chown controls, and `agentmemory remove` covers the generated runtime config. `agentmemory doctor` and startup-failure reporting were expanded accordingly.
16
17
-**Observation capture no longer grows the persisted live-stream store.** Raw and compressed observation events use `stream::send` instead of `stream::set`; the viewer still receives live updates and REST remains the history source, but `data/stream_store` no longer accumulates a write-only second and third copy of every observation.
17
18
-**Server-side capture payloads are bounded without flattening structured tool data.**`tool_input`, `tool_output`, prompts and the raw envelope are trimmed to configurable character budgets (`AGENTMEMORY_OBSERVE_*`); object fields such as `file_path` survive truncation, `0` disables an individual bound, and positive budgets below 2 are clamped to the smallest serialized string.
0 commit comments