design-playbook 0.7.0
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/LICENSE +28 -0
- package/NOTICE +37 -0
- package/README.md +143 -0
- package/commands/design-io.md +8 -0
- package/commands/ui-review.md +8 -0
- package/commands/ux-spec.md +8 -0
- package/mcp/__init__.py +0 -0
- package/mcp/_transport.py +242 -0
- package/mcp/evidence/README.md +40 -0
- package/mcp/evidence/__init__.py +0 -0
- package/mcp/evidence/server.py +450 -0
- package/mcp/evidence/test_server_stdio.py +645 -0
- package/mcp/preview/__init__.py +0 -0
- package/mcp/preview/browser.py +661 -0
- package/mcp/preview/confirm.py +255 -0
- package/mcp/preview/control.py +1293 -0
- package/mcp/preview/i18n.py +162 -0
- package/mcp/preview/server.py +126 -0
- package/mcp/preview/test_browser_control.py +663 -0
- package/mcp/preview/test_server_stdio.py +630 -0
- package/mcp/preview/test_transaction.py +436 -0
- package/mcp/preview/transaction.py +536 -0
- package/mcp/preview/util.py +19 -0
- package/mcp/test_transport.py +39 -0
- package/package.json +42 -0
- package/skills/craft-guard/SKILL.md +59 -0
- package/skills/craft-guard/references/craft.md +29 -0
- package/skills/craft-guard/references/detectors.md +124 -0
- package/skills/design-baseline/SKILL.md +134 -0
- package/skills/design-baseline/agents/openai.yaml +4 -0
- package/skills/design-baseline/references/design-template.md +73 -0
- package/skills/design-baseline/references/extraction-guidance.md +39 -0
- package/skills/design-baseline/scripts/design_baseline.py +780 -0
- package/skills/design-playbook/SKILL.md +219 -0
- package/skills/native-craft/SKILL.md +59 -0
- package/skills/native-craft/references/native-feel.md +79 -0
- package/skills/reference-intake/SKILL.md +86 -0
- package/skills/reference-intake/references/contract-template.md +82 -0
- package/skills/ui-evaluator/SKILL.md +110 -0
- package/skills/ui-evaluator/references/rubric.md +45 -0
- package/skills/ui-picker/SKILL.md +63 -0
- package/skills/ui-picker/references/components.md +31 -0
- package/skills/ui-picker/references/design.md +21 -0
- package/skills/ui-picker/references/domain.md +26 -0
- package/skills/ui-picker/references/template.md +24 -0
- package/skills/ux-spec/SKILL.md +51 -0
- package/skills/ux-spec/references/spec-template.md +43 -0
|
@@ -0,0 +1,255 @@
|
|
|
1
|
+
"""ADR-0008 floor logic + confirm/log records + prototype target resolution.
|
|
2
|
+
|
|
3
|
+
Sibling module split from server.py; behavior unchanged. Holds the
|
|
4
|
+
feedback-floor check, confirm JSON + log writers, prototype path helpers,
|
|
5
|
+
and the ``--self-check`` floor cases.
|
|
6
|
+
"""
|
|
7
|
+
from __future__ import annotations
|
|
8
|
+
|
|
9
|
+
import hashlib
|
|
10
|
+
import json
|
|
11
|
+
import secrets
|
|
12
|
+
import sys
|
|
13
|
+
import threading
|
|
14
|
+
from pathlib import Path
|
|
15
|
+
from typing import Any
|
|
16
|
+
|
|
17
|
+
from util import _now_iso
|
|
18
|
+
|
|
19
|
+
|
|
20
|
+
def prototype_html_digest(raw: bytes) -> str:
|
|
21
|
+
"""SHA-256 of prototype bytes with newlines normalized to LF.
|
|
22
|
+
|
|
23
|
+
Windows ``core.autocrlf`` rewrites working-tree bytes on checkout; a raw
|
|
24
|
+
digest then disagrees between the machine that wrote the confirm record
|
|
25
|
+
and a Linux CI runner validating the same git blob. Line-ending noise is
|
|
26
|
+
not a prototype content change for G5 integrity (issue 02 / T01).
|
|
27
|
+
|
|
28
|
+
Must stay in lockstep with ``scripts/_preview_integrity.prototype_html_digest``.
|
|
29
|
+
"""
|
|
30
|
+
return hashlib.sha256(
|
|
31
|
+
raw.replace(b"\r\n", b"\n").replace(b"\r", b"\n")
|
|
32
|
+
).hexdigest()
|
|
33
|
+
|
|
34
|
+
|
|
35
|
+
def _preview_dir_for(path: Path | None) -> Path:
|
|
36
|
+
if path is not None:
|
|
37
|
+
return path.parent
|
|
38
|
+
scratch = Path.cwd() / ".scratch" / "preview-adapter" / "preview"
|
|
39
|
+
scratch.mkdir(parents=True, exist_ok=True)
|
|
40
|
+
return scratch
|
|
41
|
+
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
def _ensure_prototype(path_arg: str | None, html: str | None, round_n: int,
|
|
45
|
+
preview_dir: Path) -> Path:
|
|
46
|
+
if path_arg:
|
|
47
|
+
p = Path(path_arg)
|
|
48
|
+
if not p.is_file():
|
|
49
|
+
raise ValueError(f"prototype path does not exist: {path_arg}")
|
|
50
|
+
return p
|
|
51
|
+
if not html:
|
|
52
|
+
raise ValueError("path or html is required")
|
|
53
|
+
preview_dir.mkdir(parents=True, exist_ok=True)
|
|
54
|
+
target = preview_dir / f"round-{round_n}.html"
|
|
55
|
+
target.write_text(html, encoding="utf-8")
|
|
56
|
+
return target
|
|
57
|
+
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
def _append_log(preview_dir: Path, *, round_n: int, report_ref: str,
|
|
61
|
+
feedback: str, aborted: bool, selected: list[str],
|
|
62
|
+
anchors: list[dict[str, Any]] | None = None,
|
|
63
|
+
floor_pass: bool | None = None,
|
|
64
|
+
floor_failure: str = "",
|
|
65
|
+
rejected: bool = False,
|
|
66
|
+
rejection: str = "") -> None:
|
|
67
|
+
preview_dir.mkdir(parents=True, exist_ok=True)
|
|
68
|
+
log_path = preview_dir / "log.md"
|
|
69
|
+
if not log_path.is_file():
|
|
70
|
+
log_path.write_text("# preview log\n", encoding="utf-8")
|
|
71
|
+
block = (
|
|
72
|
+
f"\n## round {round_n}\n"
|
|
73
|
+
f"- report_ref: {report_ref}\n"
|
|
74
|
+
f"- timestamp: {_now_iso()}\n"
|
|
75
|
+
f"- feedback: {feedback or ''}\n"
|
|
76
|
+
f"- selected: {', '.join(selected) if selected else ''}\n"
|
|
77
|
+
f"- aborted: {str(aborted).lower()}\n"
|
|
78
|
+
f"- anchors: {len(anchors or [])}\n"
|
|
79
|
+
)
|
|
80
|
+
if floor_pass is not None:
|
|
81
|
+
block += f"- floor_pass: {str(floor_pass).lower()}\n"
|
|
82
|
+
if floor_failure:
|
|
83
|
+
block += f"- floor_failure: {floor_failure}\n"
|
|
84
|
+
# LOW-4 (secure-ship-0.4.4): persist G5 fail-closed rejections (forged
|
|
85
|
+
# token / replay / round mismatch) to log.md so the event is auditable
|
|
86
|
+
# on disk, not just in the ephemeral MCP payload. Only emitted when a
|
|
87
|
+
# decision was actually rejected — a normal confirm/revise/abort leaves
|
|
88
|
+
# no rejection line.
|
|
89
|
+
if rejected:
|
|
90
|
+
block += "- rejected: true\n"
|
|
91
|
+
if rejection:
|
|
92
|
+
block += f"- rejection: {rejection}\n"
|
|
93
|
+
if anchors:
|
|
94
|
+
for i, a in enumerate(anchors, 1):
|
|
95
|
+
sel = a.get("selector") or ""
|
|
96
|
+
note = a.get("comment") or ""
|
|
97
|
+
label = a.get("label") or ""
|
|
98
|
+
block += f" - [{i}] {sel} | {label} | {note}\n"
|
|
99
|
+
with log_path.open("a", encoding="utf-8") as fh:
|
|
100
|
+
fh.write(block)
|
|
101
|
+
|
|
102
|
+
|
|
103
|
+
|
|
104
|
+
def _write_confirm(preview_dir: Path, *, round_n: int, report_ref: str,
|
|
105
|
+
selected: list[str], feedback: str,
|
|
106
|
+
confirmed: bool, floor_pass: bool,
|
|
107
|
+
prototype_html_hash: str,
|
|
108
|
+
floor_failure: str = "") -> Path:
|
|
109
|
+
record = {
|
|
110
|
+
"round": round_n,
|
|
111
|
+
"report_ref": report_ref,
|
|
112
|
+
"confirmed": confirmed,
|
|
113
|
+
"floor_pass": floor_pass,
|
|
114
|
+
"selected_options": selected,
|
|
115
|
+
"feedback": feedback,
|
|
116
|
+
"timestamp": _now_iso(),
|
|
117
|
+
"prototype_path": f"preview/round-{round_n}.html",
|
|
118
|
+
"prototype_html_hash": prototype_html_hash,
|
|
119
|
+
}
|
|
120
|
+
if floor_failure:
|
|
121
|
+
record["floor_failure"] = floor_failure
|
|
122
|
+
out = preview_dir / f"confirm-round-{round_n}.json"
|
|
123
|
+
out.write_text(json.dumps(record, ensure_ascii=False, indent=2) + "\n",
|
|
124
|
+
encoding="utf-8")
|
|
125
|
+
return out
|
|
126
|
+
|
|
127
|
+
|
|
128
|
+
|
|
129
|
+
def _check_feedback_floor(feedback: str,
|
|
130
|
+
anchors: list[dict[str, Any]]) -> tuple[bool, str]:
|
|
131
|
+
"""ADR-0008 preview feedback floor (structural, machine-checkable).
|
|
132
|
+
|
|
133
|
+
Passes when:
|
|
134
|
+
- (non-empty feedback OR >=1 anchor present) as trigger, AND
|
|
135
|
+
- every present anchor (if any) has non-empty selector AND non-empty comment.
|
|
136
|
+
|
|
137
|
+
Deliberately structural, no minimum length: short CJK feedback like
|
|
138
|
+
"太挤了" is substantive; semantic junk (ADR-0008's "安师大" case) is
|
|
139
|
+
ui-evaluator's job (G6), not the floor's.
|
|
140
|
+
Returns (floor_pass, floor_failure_reason).
|
|
141
|
+
"""
|
|
142
|
+
feedback = (feedback or "").strip()
|
|
143
|
+
trigger = bool(feedback) or bool(anchors)
|
|
144
|
+
if not trigger:
|
|
145
|
+
return False, "confirm with no substantive feedback: empty feedback and no anchor"
|
|
146
|
+
if anchors:
|
|
147
|
+
for a in anchors:
|
|
148
|
+
if not isinstance(a, dict):
|
|
149
|
+
return False, "anchor is not an object"
|
|
150
|
+
sel = str(a.get("selector") or "").strip()
|
|
151
|
+
note = str(a.get("comment") or "").strip()
|
|
152
|
+
if not sel or not note:
|
|
153
|
+
return False, (
|
|
154
|
+
"anchor missing non-empty selector and comment: "
|
|
155
|
+
f"selector={sel!r} comment={note!r}")
|
|
156
|
+
return True, ""
|
|
157
|
+
|
|
158
|
+
|
|
159
|
+
|
|
160
|
+
def _generate_decision_token() -> str:
|
|
161
|
+
"""One-time URL-safe decision token (G5 trust boundary).
|
|
162
|
+
|
|
163
|
+
Proves a POST to /decide originated from the trusted parent control bar
|
|
164
|
+
(which renders the hidden field) rather than from prototype scripts running
|
|
165
|
+
inside the sandboxed iframe, which cannot read the parent DOM. Bound to a
|
|
166
|
+
single preview round via :class:`_DecisionSession`.
|
|
167
|
+
"""
|
|
168
|
+
return secrets.token_urlsafe(32)
|
|
169
|
+
|
|
170
|
+
|
|
171
|
+
class _DecisionSession:
|
|
172
|
+
"""First-decision-wins token lock for a single preview round (G5).
|
|
173
|
+
|
|
174
|
+
``validate`` returns ``True`` only for the first POST whose token matches
|
|
175
|
+
(constant-time) AND whose round matches. Every other POST — missing token,
|
|
176
|
+
reused token, mismatched round, or wrong token — is rejected so the caller
|
|
177
|
+
can fail the decision closed. The session grants at most one valid decision.
|
|
178
|
+
"""
|
|
179
|
+
|
|
180
|
+
def __init__(self, round_n: int, token: str) -> None:
|
|
181
|
+
self.round_n = round_n
|
|
182
|
+
self._token = token
|
|
183
|
+
self._locked = False
|
|
184
|
+
self._lock = threading.Lock()
|
|
185
|
+
self.last_rejection: str = ""
|
|
186
|
+
|
|
187
|
+
@property
|
|
188
|
+
def locked(self) -> bool:
|
|
189
|
+
return self._locked
|
|
190
|
+
|
|
191
|
+
def validate(self, posted_round: int, posted_token: str | None) -> bool:
|
|
192
|
+
# LOW-1 (secure-ship-0.4.4): the check-then-set on ``_locked`` must
|
|
193
|
+
# be atomic. ThreadingHTTPServer handles each POST on its own
|
|
194
|
+
# thread, so two concurrent valid-token POSTs could both pass the
|
|
195
|
+
# ``if self._locked`` check and each consume the session. Hold the
|
|
196
|
+
# lock for the whole decision so first-decision-wins holds under
|
|
197
|
+
# real concurrency; the lock is uncontended in the single-POST
|
|
198
|
+
# happy path, so the cost is a no-op acquire/release.
|
|
199
|
+
with self._lock:
|
|
200
|
+
if not posted_token:
|
|
201
|
+
self.last_rejection = "missing"
|
|
202
|
+
return False
|
|
203
|
+
if posted_round != self.round_n:
|
|
204
|
+
self.last_rejection = "round_mismatch"
|
|
205
|
+
return False
|
|
206
|
+
if self._locked:
|
|
207
|
+
# First valid decision already consumed the session; every
|
|
208
|
+
# later POST (even with the correct token) is a replay.
|
|
209
|
+
self.last_rejection = "reuse"
|
|
210
|
+
return False
|
|
211
|
+
if not secrets.compare_digest(posted_token, self._token):
|
|
212
|
+
self.last_rejection = "invalid_token"
|
|
213
|
+
return False
|
|
214
|
+
self._locked = True
|
|
215
|
+
self.last_rejection = ""
|
|
216
|
+
return True
|
|
217
|
+
|
|
218
|
+
|
|
219
|
+
def _self_check_floor() -> None:
|
|
220
|
+
"""ADR-0008 floor branch logic self-check (ponytail: one runnable check)."""
|
|
221
|
+
cases = [
|
|
222
|
+
("empty + no anchors", "", [], False),
|
|
223
|
+
("whitespace-only feedback", " \n ", [], False),
|
|
224
|
+
("short feedback passes (structural floor)", "ok", [], True),
|
|
225
|
+
("short CJK feedback passes ('太挤了' is substantive)", "太挤了", [], True),
|
|
226
|
+
("'安师大' passes floor; semantic junk is G6's job (ADR-0008)", "安师大", [], True),
|
|
227
|
+
("longer feedback passes", "fix it", [], True),
|
|
228
|
+
("anchor with comment", "", [{"selector": "h2", "comment": "x"}], True),
|
|
229
|
+
("anchor no comment (0015 garbage)", "",
|
|
230
|
+
[{"selector": "h2", "comment": ""}], False),
|
|
231
|
+
("anchor empty selector", "",
|
|
232
|
+
[{"selector": "", "comment": "x"}], False),
|
|
233
|
+
("non-dict anchor", "", ["not-a-dict"], False),
|
|
234
|
+
("feedback + incomplete anchor still fails", "ok",
|
|
235
|
+
[{"selector": "h2", "comment": ""}], False),
|
|
236
|
+
("two anchors one incomplete fails", "",
|
|
237
|
+
[{"selector": "h2", "comment": "x"}, {"selector": "p", "comment": ""}], False),
|
|
238
|
+
("two anchors both complete passes", "",
|
|
239
|
+
[{"selector": "h2", "comment": "x"}, {"selector": "p", "comment": "y"}], True),
|
|
240
|
+
("short feedback + good anchor passes", "hi",
|
|
241
|
+
[{"selector": "h2", "comment": "x"}], True),
|
|
242
|
+
]
|
|
243
|
+
for label, fb, anc, want in cases:
|
|
244
|
+
got, _ = _check_feedback_floor(fb, anc)
|
|
245
|
+
assert got == want, f"{label}: want {want}, got {got}"
|
|
246
|
+
print("FLOOR SELF-CHECK PASSED")
|
|
247
|
+
|
|
248
|
+
|
|
249
|
+
if __name__ == "__main__":
|
|
250
|
+
if len(sys.argv) > 1 and sys.argv[1] == "--self-check":
|
|
251
|
+
_self_check_floor()
|
|
252
|
+
else:
|
|
253
|
+
raise SystemExit(
|
|
254
|
+
"usage: confirm.py --self-check (the MCP server entry is server.py)")
|
|
255
|
+
|