orchestrix-skills 0.5.0 → 0.6.0
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/package.json
CHANGED
|
@@ -2,18 +2,9 @@
|
|
|
2
2
|
|
|
3
3
|
ADR-style entries that `draft-story`, `design-architecture`, and `implement`
|
|
4
4
|
read. One entry per decision, terse, with provenance. Append-only; supersede,
|
|
5
|
-
do not rewrite.
|
|
6
|
-
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
- decision: Organize by capability (skills) wired by an orchestrator, not by role agents with handoffs.
|
|
10
|
-
- context: Role-handoff pipelines cost 30–50x in cold-start re-reads; strong models make them unnecessary.
|
|
11
|
-
- alternatives: BMAD-style role agents (rejected — handoff tax); single monolithic agent (rejected — no isolation/verification).
|
|
12
|
-
- status: accepted
|
|
13
|
-
- source: human
|
|
14
|
-
- added: 2026-06-27
|
|
15
|
-
- approved_by: dorayo
|
|
16
|
-
- ref: cc-plans/Skill编排器与最小契约设计-v0.1.md
|
|
5
|
+
do not rewrite. Seed is empty — this is honest: THIS project has made no
|
|
6
|
+
architecture decisions yet. Entries are appended by `design-architecture` (or a
|
|
7
|
+
human), never shipped as scaffold defaults.
|
|
17
8
|
|
|
18
9
|
<!--
|
|
19
10
|
Entry shape (copy for new decisions):
|
|
@@ -36,7 +36,15 @@ but always presented and approved.
|
|
|
36
36
|
help decompose into sub-projects — each gets its own spec → stories → build.
|
|
37
37
|
Don't refine details of something that should be split.
|
|
38
38
|
3. **Clarify, one question at a time.** Purpose, constraints, success criteria.
|
|
39
|
-
Prefer multiple-choice. Never batch questions
|
|
39
|
+
Prefer multiple-choice. Never batch questions while direction is still open —
|
|
40
|
+
each answer shapes the next question.
|
|
41
|
+
**Batch-questionnaire branch:** once direction is settled and what remains is
|
|
42
|
+
≥3 mutually independent detail questions, switch to ONE structured
|
|
43
|
+
questionnaire instead of asking serially — every question gets concrete
|
|
44
|
+
options plus a recommended default, so the human can answer the whole set in
|
|
45
|
+
a single pass (in Claude Code, use AskUserQuestion). Never mix modes:
|
|
46
|
+
direction questions are always serial, independent details are always
|
|
47
|
+
batched.
|
|
40
48
|
4. **Propose 2–3 approaches** with trade-offs; lead with your recommendation and
|
|
41
49
|
why.
|
|
42
50
|
5. **Present the design in sections**, scaled to complexity; confirm each
|