Add auto refine review hook

packages/coding-agent/CHANGELOG.md

@@ -2,6 +2,8 @@
 
 ## [Unreleased]
 
+- Added an opt-in auto-refine review hook that can ask whether `/refine` should run after turn intervals or compaction checkpoints.
+
 ## [0.2.4] - 2026-07-01
 
 - Changed the agents view to list only sessions the daemon is actively holding, and stopped the daemon from auto-restoring on-disk sessions on startup, so a restarted daemon no longer surfaces a wall of weeks-old sessions; sessions come back via `/resume` or `--resume` ([#295](https://github.com/PrimeIntellect-ai/prime-agent/issues/295)).

packages/coding-agent/docs/kernel-and-rlm-recursion.md

@@ -118,9 +118,12 @@ rlm.harness.record_refinement(
 print(rlm.harness.overview())
 ```
 
-The store writes `harness_state.json` in the global agent harness directory
-(`RLM_HARNESS_STATE_DIR`, e.g. `~/.prime/agent/harness/`), so learned state is
-shared across sessions. Because the long-lived kernel and the host `/refine`
+The store writes `harness_state.json` in the session-local harness directory by
+default (`RLM_HARNESS_STATE_DIR`, falling back to `RLM_SESSION_DIR/harness`), so
+learned state stays with the session. Explicitly global edits go to the global
+agent harness directory (`RLM_GLOBAL_HARNESS_STATE_DIR`, e.g.
+`~/.prime/agent/harness/`), which is shared across sessions. Because the
+long-lived kernel and the host `/refine`
 command write the same file from separate processes, the kernel-side store
 reloads the file whenever its on-disk mtime changes before reading or mutating,
 so concurrent host edits are merged rather than clobbered. It is intentionally a

packages/coding-agent/docs/refine-verification-log.md

@@ -5,10 +5,14 @@ harness features. It is intentionally artifact-oriented so later benchmark runs
 can replay the same surfaces.
 
 Current design note: refined prompt notes, memories, skills, subagent specs, and
-refinement events are global by default under the agent harness directory, for
-example `~/.prime/agent/harness/harness_state.json`. Session JSONL entries still
+refinement events are session-local by default under the session harness
+directory (`RLM_HARNESS_STATE_DIR`, falling back to `RLM_SESSION_DIR/harness`);
+explicitly global refinement writes the global agent harness directory
+(`RLM_GLOBAL_HARNESS_STATE_DIR`, for example
+`~/.prime/agent/harness/harness_state.json`). Session JSONL entries still
 record refinement results for auditability and rollback evidence. A compact
-overview of the global harness state is injected into the default system prompt
+overview of the merged global and local harness state is injected into the
+default system prompt
 so the agent can use learned state without first calling `rlm.harness.overview()`.
 The model-facing `rlm.harness` API uses explicit `create_*`, `update_*`, and
 `delete_*` calls for memory, skill, subagent, and prompt-note entries.
@@ -204,3 +208,10 @@ Covered validation and recovery cases:
 - Python runtime default backing store through `RLM_HARNESS_STATE_DIR`.
 - Python runtime unknown-kind rejection for `upsert`, `get`, `delete`, and
   `list`.
+
+## 2026-07-01 correction
+
+When the entries above were recorded, `RLM_HARNESS_STATE_DIR` pointed at the
+global harness directory. Since local-by-default refinement landed, the global
+directory is `RLM_GLOBAL_HARNESS_STATE_DIR` and `RLM_HARNESS_STATE_DIR` is the
+session-local store (falling back to `RLM_SESSION_DIR/harness`).