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
fix(classifier): drop --bare, gate hook recursion via TJ_IN_CLASSIFIER (v0.2.9)
claude --bare ignores ~/.claude/.credentials.json and demands
ANTHROPIC_API_KEY. Subscription-auth users (Pro/Max — most of the
audience) hit "Not logged in" on every classification, so the
journal stayed empty in production. claude-memory-0kk.
We previously needed --bare to stop the classifier path from
spawning a `claude -p` that re-reads settings.json and re-fires our
hooks recursively. Replace that mechanism with an explicit sentinel
env var:
- ClaudeCliClassifier.classify() drops --bare from argv and sets
TJ_IN_CLASSIFIER=1 on the spawned process.
- IngestHook handler returns Ok(()) immediately when it sees
TJ_IN_CLASSIFIER in env. Recursion broken without touching auth.
One regression test (ingest_hook_short_circuits_when_in_classifier_
env_set) plus the existing 40 ingest-hook tests stay green.
Trade-off: without --bare the inner claude loads CLAUDE.md / skills
on the first call per 5-minute prompt-cache window (one-time ~60K
cache_creation tokens; subsequent calls hit cache and stay cheap).
The classifier prompt explicitly demands JSON-only output, so model
compliance survives the larger system prompt.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
"description": "Append-only journal of AI-coding task reasoning chains. Captures hypotheses, decisions, rejections, evidence — renders compact resume packs so an agent can pick up a 2-week-old task with full context.",
Copy file name to clipboardExpand all lines: plugin/.claude-plugin/plugin.json
+1-1Lines changed: 1 addition & 1 deletion
Original file line number
Diff line number
Diff line change
@@ -1,6 +1,6 @@
1
1
{
2
2
"name": "task-journal",
3
-
"version": "0.2.8",
3
+
"version": "0.2.9",
4
4
"description": "Append-only journal of AI-coding task reasoning chains: hypotheses, decisions, rejections, evidence. Renders compact resume packs so an agent can pick up a 2-week-old task with full context.",
0 commit comments