Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -8,9 +8,12 @@ Please choose versions by [Semantic Versioning](http://semver.org/).
* MINOR version when you add functionality in a backwards-compatible manner, and
* PATCH version when you make backwards-compatible bug fixes.

## v0.169.2
## Unreleased

- fix: `/vault-cli:sync-progress` § Phase 2 reaches its `already synced — no writes this run` short-circuit for an `in_progress` anchor, instead of reporting failure for a record it can verify as present. The short-circuit added in `v0.168.6` was gated *"When a completion was established above"*, so the state the command is most often run in — an anchor that is still open — fell past it to `No task completion or PR detected. Use /update instead for in-progress work.` and STOP. Measured 2026-10-15: the operator ran the command **four times in one session** against an `in_progress` anchor whose daily-note entry already existed; all four stopped at the same line, each costing a full Phase 1–2 sweep (vault config read, session-transcript scan, git probes) and producing an identical report. The gate now reads *"whenever a completion was established above **or** the anchor is `in_progress`"*, and an explicit `in_progress` arm precedes the STOP bullet so the representation check runs first. Still keyed on the representation check rather than on the status alone: an `in_progress` anchor with **no** daily-note entry falls through to the STOP line unchanged. Out of scope: the Phase 1–2 sweep itself (the short-circuit still runs after the completion analysis, so the sweep is not removed — the false failure report and the operator's reason to re-run are) and the close-family loop in `/vault-cli:session-close`.

## v0.169.2

- fix: `/vault-cli:create-task` and its `task-creator` agent no longer fall back silently to the configured default vault. `commands/create-task.md` documented `--vault` as optional with an implicit default, and `agents/task-creator.md` stated it outright (`Defaults: MODE = interactive, VAULT = configured default vault`) — so a task belonging to one vault was placed in another by inference, and the silent fallback is a **third** vault again, wrong for work belonging to any other. Both now resolve the vault from `--vault` or from the cwd, and **ask** when neither determines it — the convention `commands/reflect.md` (*"If cwd is not inside any vault path, ask the user which vault to write to"*) and `agents/graph-auditor.md` (*"ask user which vault to audit; do not guess"*) already carry, so this brings the pair in line with their siblings rather than adding a new rule. Observed 2026-10-15: a task belonging to `seibert-personal` was created in `seibert-brogrammers` by inference and had to be deleted and re-created, costing two `task-creator` runs (~105k tokens).

## v0.169.1
Expand Down
Loading