ai-wiki-toolkit-linux-arm64 0.1.37 → 0.1.38

package/README.md

@@ -1,7 +1,7 @@
 # ai-wiki-toolkit-linux-arm64
 
 This package contains the `aiwiki-toolkit` executable for `linux-arm64-glibc`.
-It is published as the platform-specific binary package for `ai-wiki-toolkit` `0.1.37`.
+It is published as the platform-specific binary package for `ai-wiki-toolkit` `0.1.38`.
 Most users should install `ai-wiki-toolkit` instead of using this package directly.
 
 ---
@@ -334,6 +334,16 @@ non-interactive shells, pass `--handle your-name` or set `AIWIKI_TOOLKIT_HANDLE`
 
 The tool works best when `git user.name` and `git user.email` are configured first.
 
+For stronger trial/error capture, configure your agent runner to call the post-turn capture command
+after completed AI Wiki write-back turns:
+
+```bash
+aiwiki-toolkit source-incident capture-post-turn --apply
+```
+
+This hook is recommended when your runner supports it, but `install` does not enable hooks
+automatically or mutate agent runtime configuration.
+
 If package-owned repo-local skill files already exist under `.agents/skills/ai-wiki-*`, the installer refreshes them from the current package so new workflow and footer contracts propagate on upgrade. Because these files live in git, local customizations remain visible in `git diff` after running `install`.
 
 `init` remains as a backward-compatible alias for `install`. The actual scaffold creation does not happen at package install time; it happens when you run `aiwiki-toolkit install` or `aiwiki-toolkit init` inside a git repository.
@@ -387,6 +397,24 @@ aiwiki-toolkit record-reuse \
 
 Diagnostics and `eval impact discover` report this as `source_incident_timing`. Treat it as source active-turn context for research, not as exact human time saved or a formal no-AI-wiki baseline.
 
+For older memories where the original `record-reuse` event did not include source timing, backfill a separate evidence ledger from local Codex write-back footers:
+
+```bash
+aiwiki-toolkit source-incident backfill-writeback \
+  --writeback-path ai-wiki/problems/retry-loop.md \
+  --apply
+```
+
+This scans local `~/.codex/sessions` for the first `AI Wiki Write-Back Path:` footer in a session whose `cwd` matches the current repository, counts active `task_complete.duration_ms` plus timed `turn_aborted.duration_ms` rows from the current user task start through that first write-back turn, and appends the result to `ai-wiki/metrics/source-incidents/<handle>.jsonl`. It does not mutate historical reuse events. Omit `--apply` for a dry run, or pass `--doc-id`/`--writeback-path` to scope the backfill.
+
+For a post-turn hook or wrapper, capture only the latest completed write-back turn for the current repo:
+
+```bash
+aiwiki-toolkit source-incident capture-post-turn --apply
+```
+
+This command is idempotent. If the latest write-back was already captured, it reports `skipped_existing` instead of appending another row. A runner that knows the Codex session id can pass `--session-id <id>` to avoid scanning all local session files.
+
 Only record user-owned AI wiki knowledge docs with `record-reuse`.
 
 Managed control-plane docs under `_toolkit/**` should still be cited in user-facing notes when they affect behavior, but they should not be logged as knowledge-reuse events.
@@ -436,10 +464,11 @@ aiwiki-toolkit record-reuse-check \
 
 This appends to the user-owned `ai-wiki/metrics/task-checks/<handle>.jsonl` shard and refreshes the handle-scoped generated views under `ai-wiki/_toolkit/metrics/by-handle/<handle>/`. The installer ignores both the shard and the generated aggregate views by default so these telemetry updates stay local.
 
-Both metrics logs are sharded by handle under:
+The local metrics logs are sharded by handle under:
 
 - `ai-wiki/metrics/reuse-events/<handle>.jsonl`
 - `ai-wiki/metrics/task-checks/<handle>.jsonl`
+- `ai-wiki/metrics/source-incidents/<handle>.jsonl`
 
 These logs are intended as local telemetry by default, not merge-heavy source files.
 
@@ -463,7 +492,9 @@ Use `--focus trial-error` to generate a focused trial/error reduction report fro
 AI wiki evidence. It summarizes material effects such as `avoided_retry`,
 `blocked_wrong_path`, `changed_plan`, and `faster_resolution`, separates missed or repeated issue
 signals from unproven wiki use, and lists replay candidates that still need source incident
-artifacts before becoming formal impact-eval families.
+artifacts before becoming formal impact-eval families. Source incident timing can come either from
+new `record-reuse` provenance fields or from the separate `source-incident backfill-writeback`
+ledger.
 
 To turn diagnostics and handle-local drafts into a human-reviewable consolidation queue:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-wiki-toolkit-linux-arm64",
-  "version": "0.1.37",
+  "version": "0.1.38",
   "description": "Platform binary package for ai-wiki-toolkit (linux-arm64-glibc).",
   "license": "MIT",
   "homepage": "https://github.com/BochengYin/ai-wiki-toolkit#readme",