@curdx/flow 7.1.13 → 7.1.15
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +28 -0
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -2,6 +2,34 @@
|
|
|
2
2
|
|
|
3
3
|
All notable changes to `@curdx/flow` are documented here. Format follows [Keep a Changelog](https://keepachangelog.com/) and the project follows [Semantic Versioning](https://semver.org/).
|
|
4
4
|
|
|
5
|
+
## 7.1.15 — 2026-05-11
|
|
6
|
+
|
|
7
|
+
### Added
|
|
8
|
+
|
|
9
|
+
- **Deterministic AutoPolicy execution classifier.** New hook utility `hooks/scripts/lib/auto-policy.mjs` classifies each goal into XS/S/M/L/XL size, risk, execution mode, task target range, review cadence, verification level, subagent policy, Stop-hook behavior, and iteration caps before expensive model work starts.
|
|
10
|
+
- **Runtime and schema support for persisted `autoPolicy`.** `.curdx-state.json` now stores the selected policy, schema validation accepts `autoPolicy`, and shared hook types expose the policy contract.
|
|
11
|
+
- **Regression tests for policy behavior and workflow drift.** New hook tests lock classifier decisions and Stop-hook policy behavior; runner tests prevent task sizing docs from drifting back to 40+ task defaults.
|
|
12
|
+
|
|
13
|
+
### Changed
|
|
14
|
+
|
|
15
|
+
- **Specs now default to automatic strength selection instead of asking users to choose fast/deep or fine/coarse.** `/curdx-flow:start`, `/curdx-flow:new`, `/curdx-flow:tasks`, `/curdx-flow:implement`, help, core references, and task-planner guidance all treat `autoPolicy` as the source of truth while still allowing explicit `--mode`, `--tasks-size`, and `--review` overrides.
|
|
16
|
+
- **Task planning is now bounded around vertical slices.** Top-level tasks keep test/reproduce, implementation, verification, and commit together; normal specs target 3-7 or 5-12 tasks, and oversized requests route to triage instead of inflating one spec.
|
|
17
|
+
- **Stop-hook continuation is shorter when policy allows it.** Medium and larger specs use a compact continuation prompt; XS/S specs can disable automatic Stop-hook continuation entirely to avoid token-heavy loop prompts.
|
|
18
|
+
|
|
19
|
+
## 7.1.14 — 2026-05-10
|
|
20
|
+
|
|
21
|
+
### Fixed
|
|
22
|
+
|
|
23
|
+
- **Fresh spec state now consistently uses the bounded `maxGlobalIterations: 30` default.** `/curdx-flow:start`, `/curdx-flow:new`, quick mode, and the execution-state hook fallback now agree with the implement skill's cost-runaway contract. Existing legacy state files that already store `100` remain compatible.
|
|
24
|
+
- **Agent Teams are no longer documented as the stable default anywhere in the workflow path.** Research, triage, scanner resume guidance, and core delegation rules now use direct `Task(...)` dispatch as the baseline, with Agent Teams only as an optional experimental overlay when enabled and available.
|
|
25
|
+
- **Public entrypoint skill tool grants are narrower.** Shipped `/curdx-flow:*` skills no longer use `allowed-tools: "*"`, reducing accidental broad permission grants while preserving the tools each workflow actually needs.
|
|
26
|
+
- **Support skills are hidden from the slash menu but remain model-invocable background guidance.** `spec-workflow`, `curdx-core`, `communication-style`, `interview-framework`, and `verification-before-completion` now have trigger-focused descriptions and explicit background-skill metadata.
|
|
27
|
+
- **Help and cost references now expose the current iteration-cap interface.** `/curdx-flow:help` documents both `--max-task-iterations` and `--max-global-iterations`; cost docs use the real `/curdx-flow:implement` invocation form.
|
|
28
|
+
|
|
29
|
+
### Added
|
|
30
|
+
|
|
31
|
+
- **Runner guards for runtime-contract drift.** Manifest integrity tests now lock bounded iteration defaults, forbid wildcard grants on public entrypoints, require support skills to stay hidden/background-invocable, reject old "research team as default" wording, and verify help documents the global iteration cap.
|
|
32
|
+
|
|
5
33
|
## 7.1.13 — 2026-05-10
|
|
6
34
|
|
|
7
35
|
### Changed
|