oh-my-customcode 1.1.78 → 1.1.80
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/dist/cli/index.js +1 -1
- package/dist/index.js +1 -1
- package/package.json +1 -1
- package/templates/.claude/rules/MUST-agent-design.md +2 -0
- package/templates/.claude/skills/agora/scripts/anonymize.sh +166 -13
- package/templates/guides/claude-code/15-version-compatibility.md +3 -3
- package/templates/manifest.json +1 -1
package/dist/cli/index.js
CHANGED
package/dist/index.js
CHANGED
package/package.json
CHANGED
|
@@ -118,6 +118,8 @@ Fable 5는 high effort가 기본값이며, `xhigh`는 capability-sensitive 작
|
|
|
118
118
|
|
|
119
119
|
<!-- DETAIL: effort frontmatter CC version notes (historical)
|
|
120
120
|
> **v2.1.267+**: `effort:` 프론트매터 관련 결함·신규 상한 3건 — (a) 커스텀 커맨드·스킬·서브에이전트의 `effort:` 프론트매터가, 기본 effort가 여전히 고정된 모델(Opus 4.7, Opus 4.8, Fable 5)에서 **무시**되던 결함이 수정되었습니다. 즉 267 이전에는 이 파일의 "스킬 `effort`가 에이전트 `effort`보다 우선한다"는 서술과 위 「Fable 5 Effort 전략」의 `effort` 명시 지침이 Fable 5 / Opus 4.8 에이전트에서 **런타임 효과가 없었습니다** — 프론트매터 effort는 실제 실행된 effort의 증거가 아니었습니다. (b) 신규 `maxEffortLevel` 설정(최상위 또는 `modelSettings` 하위 모델별)이 모든 provider에서 effort 레벨 상한을 강제합니다 — 사용자는 여전히 더 낮은 레벨을 선택할 수 있습니다. 이는 프론트매터 `effort`보다 **상위에 위치하는 설정 레벨 상한**이므로, `xhigh`를 선언한 에이전트도 상한이 설정돼 있으면 그 상한에서 실행됩니다(프론트매터로 유추하지 말고 실효 effort를 확인). (c) `/model opusplan[1m]`이 "Model not found"로 거부되던 결함이 265에서 수정되어, `/model` 명령에서 이 표기가 이제 수용됩니다 — 프론트매터 `model: opusplan[1m]` 경로는 릴리즈 노트가 언급하지 않으므로 미실측입니다.
|
|
121
|
+
>
|
|
122
|
+
> **후속 (v2.1.280+)**: CHANGELOG 원문 — "Changed Opus 4.7, Opus 4.8 and Fable 5 to stop holding their launch-default effort over `/effort` in `-p` or the Agent SDK, a project, managed or `--settings` `effortLevel`, or a per-model level." 이는 위 (a) 프론트매터 결함(267에서 이미 수정됨)과는 별개로, `-p`/Agent SDK의 `/effort`·프로젝트/managed/`--settings`의 `effortLevel`·모델별 레벨이 launch-default effort에 눌리던 문제를 해소한 후속 조치입니다 — 프론트매터·스킬/에이전트 effort 우선순위에 대한 영향은 릴리즈 노트가 언급하지 않아 미실측이며, 위 (b) `maxEffortLevel` 상한 원칙은 이 후속 조치와 무관하게 그대로 유효합니다.
|
|
121
123
|
-->
|
|
122
124
|
|
|
123
125
|
<!-- DETAIL: Mythos 5 (non-GA, not registered — reference only)
|
|
@@ -34,20 +34,22 @@ hash_int() {
|
|
|
34
34
|
}
|
|
35
35
|
|
|
36
36
|
# ---------------------------------------------------------------------------
|
|
37
|
-
#
|
|
37
|
+
# label_permutation <r-list> <vendor_id>... — apply a precomputed Fisher-Yates
|
|
38
|
+
# swap sequence and emit A/B/C JSON. <r-list> is a single space-joined string
|
|
39
|
+
# of decimal integers, one per swap step, ordered i=n-1..1 (the same order the
|
|
40
|
+
# loop below walks) — a STRING, not a bash array: a 1-vendor caller produces
|
|
41
|
+
# an EMPTY r-list, and under `set -u` a zero-element bash array is unbound on
|
|
42
|
+
# access (bash 3.2 quirk), whereas an empty string has no such trap.
|
|
38
43
|
# ---------------------------------------------------------------------------
|
|
39
|
-
|
|
40
|
-
local
|
|
41
|
-
if [ "$#" -eq 0 ]; then
|
|
42
|
-
printf 'anonymize.sh: shuffle_labels needs at least one vendor\n' >&2
|
|
43
|
-
return 64
|
|
44
|
-
fi
|
|
45
|
-
|
|
44
|
+
label_permutation() {
|
|
45
|
+
local r_list="$1"; shift
|
|
46
46
|
local vendors=("$@")
|
|
47
47
|
local n=${#vendors[@]}
|
|
48
48
|
local i j r tmp
|
|
49
|
+
# shellcheck disable=SC2086 # intentional word-split of the r-list string
|
|
50
|
+
set -- $r_list
|
|
49
51
|
for (( i = n - 1; i > 0; i-- )); do
|
|
50
|
-
r
|
|
52
|
+
r="$1"; shift
|
|
51
53
|
j=$(( r % (i + 1) ))
|
|
52
54
|
tmp="${vendors[$i]}"
|
|
53
55
|
vendors[$i]="${vendors[$j]}"
|
|
@@ -64,6 +66,160 @@ shuffle_labels() {
|
|
|
64
66
|
printf '%s\n' "$out"
|
|
65
67
|
}
|
|
66
68
|
|
|
69
|
+
# ---------------------------------------------------------------------------
|
|
70
|
+
# shuffle_labels <seed> <vendor_id>... — seeded Fisher-Yates, then A/B/C in order.
|
|
71
|
+
# ---------------------------------------------------------------------------
|
|
72
|
+
shuffle_labels() {
|
|
73
|
+
local seed="$1"; shift
|
|
74
|
+
if [ "$#" -eq 0 ]; then
|
|
75
|
+
printf 'anonymize.sh: shuffle_labels needs at least one vendor\n' >&2
|
|
76
|
+
return 64
|
|
77
|
+
fi
|
|
78
|
+
|
|
79
|
+
local n=$#
|
|
80
|
+
local i rs=''
|
|
81
|
+
for (( i = n - 1; i > 0; i-- )); do
|
|
82
|
+
rs+="$(hash_int "$seed" "$i") "
|
|
83
|
+
done
|
|
84
|
+
|
|
85
|
+
label_permutation "$rs" "$@"
|
|
86
|
+
}
|
|
87
|
+
|
|
88
|
+
# ---------------------------------------------------------------------------
|
|
89
|
+
# hash_hex_stream — batch form of hash_int's hex step: reads "<seed>:<counter>"
|
|
90
|
+
# lines on stdin, emits one 8-hex-char digest per line. `shasum` itself is a
|
|
91
|
+
# perl script wrapping Digest::SHA (`head -1 "$(which shasum)"` shows the
|
|
92
|
+
# shebang `#!/usr/bin/perl` — an ABSOLUTE path), but this function resolves
|
|
93
|
+
# `perl` from PATH instead of that fixed interpreter; the two usually
|
|
94
|
+
# coincide but are not guaranteed identical. What "no new dependency" means
|
|
95
|
+
# precisely: Digest::SHA is already required because shasum needs it, so
|
|
96
|
+
# calling perl directly here trades N subprocess forks — one `shasum`
|
|
97
|
+
# pipeline per Fisher-Yates swap — for one perl process handling every swap
|
|
98
|
+
# of every seed in a `--shuffle-many` batch (spec #1724).
|
|
99
|
+
# ---------------------------------------------------------------------------
|
|
100
|
+
hash_hex_stream() {
|
|
101
|
+
perl -MDigest::SHA=sha256_hex -lne 'print substr(sha256_hex($_), 0, 8)'
|
|
102
|
+
}
|
|
103
|
+
|
|
104
|
+
# ---------------------------------------------------------------------------
|
|
105
|
+
# shuffle_many <count> <vendor_id>... — batch form of calling
|
|
106
|
+
# `shuffle_labels "agora-shuffle-$k" <vendor_id>...` for k=1..count and
|
|
107
|
+
# concatenating the output, but with ALL hashing for the whole batch done by
|
|
108
|
+
# a SINGLE hash_hex_stream invocation instead of one `shasum` pipeline per
|
|
109
|
+
# swap per k (#1724 — CI timeout on 600 sequential shuffles). Emits one JSON
|
|
110
|
+
# map per line, same as the loop it replaces.
|
|
111
|
+
# ---------------------------------------------------------------------------
|
|
112
|
+
shuffle_many() {
|
|
113
|
+
local count="$1"; shift
|
|
114
|
+
|
|
115
|
+
# Validate count as a non-negative DECIMAL integer, rejecting anything a
|
|
116
|
+
# bare `[ "$count" -ge 1 ]` would either silently accept-as-wrong or blow
|
|
117
|
+
# up on later (measured against the original script, a70f6f47): "abc",
|
|
118
|
+
# "3.0", "0x3" and "1+1" all made `[` itself fail with "integer expression
|
|
119
|
+
# expected" while `shuffle_many` had ALREADY returned rc 0 with no output
|
|
120
|
+
# (the `||` short-circuits `[`'s failure into the same path as count=0);
|
|
121
|
+
# "08" reached the arithmetic `for` below and aborted there instead, with
|
|
122
|
+
# "value too great for base", because bash arithmetic context reads a
|
|
123
|
+
# leading-zero numeral as OCTAL and "8" is not a valid octal digit. Every
|
|
124
|
+
# one of those is a silent-or-confusing failure for a caller that passed a
|
|
125
|
+
# non-integer; reject them all up front with one explicit message instead.
|
|
126
|
+
#
|
|
127
|
+
# Design choices, both intentionally more conservative than the original:
|
|
128
|
+
# - Negative counts ("-3"): the original returned rc 0 with no output,
|
|
129
|
+
# identical to a real count of 0 — indistinguishable from "the batch
|
|
130
|
+
# was empty on purpose". Reject instead so a negative count can never
|
|
131
|
+
# be mistaken for a legitimate empty result (spec #1724 review M1).
|
|
132
|
+
# - Leading zeros ("08", "00"): reject rather than normalize (e.g. by
|
|
133
|
+
# stripping to "8"), because normalization would let two different
|
|
134
|
+
# caller-supplied strings silently produce the same batch size — an
|
|
135
|
+
# error is more honest than a silent reinterpretation. "0" itself
|
|
136
|
+
# (the literal zero, not zero-with-padding) stays valid and unchanged:
|
|
137
|
+
# rc 0, no output, exactly as before.
|
|
138
|
+
case "$count" in
|
|
139
|
+
0) ;;
|
|
140
|
+
*[!0-9]*|'')
|
|
141
|
+
printf 'anonymize.sh: --shuffle-many count must be a non-negative integer\n' >&2
|
|
142
|
+
return 64
|
|
143
|
+
;;
|
|
144
|
+
0*)
|
|
145
|
+
printf 'anonymize.sh: --shuffle-many count must be a non-negative integer\n' >&2
|
|
146
|
+
return 64
|
|
147
|
+
;;
|
|
148
|
+
esac
|
|
149
|
+
|
|
150
|
+
[ "$count" -ge 1 ] || return 0
|
|
151
|
+
|
|
152
|
+
if [ "$#" -eq 0 ]; then
|
|
153
|
+
# Zero vendors: reuse shuffle_labels for its existing error message and
|
|
154
|
+
# exit code (64) — under `set -e` this aborts the whole script on the
|
|
155
|
+
# first call, exactly as the original per-k loop did on its first
|
|
156
|
+
# iteration, regardless of how large `count` is.
|
|
157
|
+
shuffle_labels "agora-shuffle-1" "$@"
|
|
158
|
+
return
|
|
159
|
+
fi
|
|
160
|
+
|
|
161
|
+
local n=$#
|
|
162
|
+
local per_row=$(( n - 1 ))
|
|
163
|
+
local k
|
|
164
|
+
|
|
165
|
+
if [ "$per_row" -le 0 ]; then
|
|
166
|
+
# Single-vendor case: no Fisher-Yates swaps needed, so no hashing at all —
|
|
167
|
+
# the same short-circuit label_permutation already takes for an empty
|
|
168
|
+
# r-list, just repeated `count` times.
|
|
169
|
+
for (( k = 1; k <= count; k++ )); do
|
|
170
|
+
label_permutation '' "$@"
|
|
171
|
+
done
|
|
172
|
+
return
|
|
173
|
+
fi
|
|
174
|
+
|
|
175
|
+
# Build every "<seed>:<counter>" line for the whole batch up front, in the
|
|
176
|
+
# same nested order (k outer, i=n-1..1 inner) that the per-row consumption
|
|
177
|
+
# loop below expects, then hash them all in one perl process.
|
|
178
|
+
local i lines=''
|
|
179
|
+
for (( k = 1; k <= count; k++ )); do
|
|
180
|
+
for (( i = n - 1; i > 0; i-- )); do
|
|
181
|
+
lines+="agora-shuffle-$k:$i"$'\n'
|
|
182
|
+
done
|
|
183
|
+
done
|
|
184
|
+
|
|
185
|
+
local hexes
|
|
186
|
+
hexes=$(printf '%s' "$lines" | hash_hex_stream)
|
|
187
|
+
|
|
188
|
+
# hash_hex_stream runs under a pipeline, and `set -euo pipefail` only
|
|
189
|
+
# catches a non-zero EXIT from perl — it says nothing about the perl
|
|
190
|
+
# process emitting fewer lines than it was fed while still exiting 0
|
|
191
|
+
# (e.g. a truncating wrapper ahead of the real perl on PATH). A short
|
|
192
|
+
# stream here is silent data loss: label_permutation would just get
|
|
193
|
+
# called fewer times than `count`, with no error at all. Count the hex
|
|
194
|
+
# lines actually consumed and require BOTH that the total matches
|
|
195
|
+
# count*per_row (spec #1724 review L1) AND that the consume loop below
|
|
196
|
+
# ends with `seen -eq 0` — i.e. the last group closed exactly on a
|
|
197
|
+
# `per_row` boundary rather than being cut off mid-group. The second
|
|
198
|
+
# check is redundant with the first in every case this function can
|
|
199
|
+
# reach (count*per_row is itself a multiple of per_row, so a short total
|
|
200
|
+
# is caught either way), but it is cheap and it directly verifies the
|
|
201
|
+
# invariant the consume loop below depends on, rather than trusting the
|
|
202
|
+
# arithmetic that implies it.
|
|
203
|
+
local hex rs='' seen=0 total=0
|
|
204
|
+
while IFS= read -r hex; do
|
|
205
|
+
rs+="$(( 16#$hex )) "
|
|
206
|
+
seen=$(( seen + 1 ))
|
|
207
|
+
total=$(( total + 1 ))
|
|
208
|
+
if [ "$seen" -eq "$per_row" ]; then
|
|
209
|
+
label_permutation "$rs" "$@"
|
|
210
|
+
rs=''
|
|
211
|
+
seen=0
|
|
212
|
+
fi
|
|
213
|
+
done <<< "$hexes"
|
|
214
|
+
|
|
215
|
+
local expected=$(( count * per_row ))
|
|
216
|
+
if [ "$total" -ne "$expected" ] || [ "$seen" -ne 0 ]; then
|
|
217
|
+
printf 'anonymize.sh: --shuffle-many hash stream truncated: got %s hex digest(s), expected %s\n' \
|
|
218
|
+
"$total" "$expected" >&2
|
|
219
|
+
return 70
|
|
220
|
+
fi
|
|
221
|
+
}
|
|
222
|
+
|
|
67
223
|
# ---------------------------------------------------------------------------
|
|
68
224
|
# Fingerprint guard (spec §12-(1)). Case-insensitive; a hit aborts the session.
|
|
69
225
|
# Three tiers, because "vendor name" and "ordinary English word" overlap:
|
|
@@ -464,10 +620,7 @@ main() {
|
|
|
464
620
|
--shuffle-many)
|
|
465
621
|
shift
|
|
466
622
|
local count="$1"; shift
|
|
467
|
-
|
|
468
|
-
for (( k = 1; k <= count; k++ )); do
|
|
469
|
-
shuffle_labels "agora-shuffle-$k" "$@"
|
|
470
|
-
done
|
|
623
|
+
shuffle_many "$count" "$@"
|
|
471
624
|
;;
|
|
472
625
|
--build)
|
|
473
626
|
shift
|
|
@@ -1320,13 +1320,13 @@ Opus 4.8에서 thinking blocks가 수정되어 API 오류가 발생하던 버그
|
|
|
1320
1320
|
### 모델 · effort
|
|
1321
1321
|
|
|
1322
1322
|
- CHANGELOG 원문: "Added Claude Opus 5.5 (`claude-opus-5-5`), now the default Opus model — 1M context, $4/$20 per Mtok with $0.20/Mtok cache reads"
|
|
1323
|
-
(280) Opus 5.5가 기본 Opus 모델이 되었으므로, R006 Model Specification Tier 2 표에 `claude-opus-5-5` 전체 ID를 추가할지 검토가 필요합니다 — 아래
|
|
1323
|
+
(280) Opus 5.5가 기본 Opus 모델이 되었으므로, R006 Model Specification Tier 2 표에 `claude-opus-5-5` 전체 ID를 추가할지 검토가 필요합니다 — 아래 Action items 참조.
|
|
1324
1324
|
- CHANGELOG 원문: "Changed the default model on Pro and Team Standard plans from Sonnet to Opus, matching Max, Team Premium, and Enterprise"
|
|
1325
1325
|
(280) Pro/Team Standard 플랜의 기본 모델이 Sonnet에서 Opus로 바뀌었으므로, 해당 플랜을 쓰는 세션은 R005 비용 인식 관점에서 기본값 자체가 더 비싸진 상태로 시작됩니다.
|
|
1326
1326
|
- CHANGELOG 원문: "Changed an effort level saved before `/effort` became per-model to no longer apply to newly released models such as Opus 5.5; they start at their default until you pick a level"
|
|
1327
1327
|
(280) Opus 5.5처럼 새로 출시된 모델은 과거 저장된 effort 값을 물려받지 않고 기본 effort로 시작하므로, 이 저장소의 effort 관련 지침을 새 모델에 그대로 적용하기 전 실측이 필요합니다.
|
|
1328
1328
|
- CHANGELOG 원문: "Changed Opus 4.7, Opus 4.8 and Fable 5 to stop holding their launch-default effort over `/effort` in `-p` or the Agent SDK, a project, managed or `--settings` `effortLevel`, or a per-model level"
|
|
1329
|
-
(280) Opus 4.7·4.8·Fable 5
|
|
1329
|
+
(280) Opus 4.7·4.8·Fable 5는 `-p`나 Agent SDK에서 쓰는 `/effort`, 프로젝트·managed·`--settings` `effortLevel`, 모델별 레벨에 대해 launch-default effort를 더 이상 우선시하지 않게 되었습니다. R006이 v2.1.267 노트에서 기록한 것은 이 세 모델의 커스텀 커맨드·스킬·서브에이전트 `effort:` 프론트매터 무시 결함이 267에서 **수정됐다**는 사실이며, 280은 그 수정을 세션·설정 경로로 확장한 후속 변경입니다 — 아래 Action items 참조.
|
|
1330
1330
|
|
|
1331
1331
|
### MCP
|
|
1332
1332
|
|
|
@@ -1392,7 +1392,7 @@ Opus 4.8에서 thinking blocks가 수정되어 API 오류가 발생하던 버그
|
|
|
1392
1392
|
기타 91건 — 이 저장소 비해당(VSCode/Claude Code on the web/Claude Tag/Code Review 전용 29건, Windows·self-hosted runner 전용 4건 포함, 나머지는 UI 다이얼로그·마우스·키바인딩 폴리시, Artifact 도구 세부사항, 마켓플레이스 커밋 추적 등 저빈도 항목).
|
|
1393
1393
|
|
|
1394
1394
|
**Action items**:
|
|
1395
|
-
- R006 Model Specification Tier 2 표에 `claude-opus-5-5` 행을 v1.1.77에서 추가 완료. v2.1.267 DETAIL 노트(Opus 4.7·4.8·Fable 5 effort 무시
|
|
1395
|
+
- R006 Model Specification Tier 2 표에 `claude-opus-5-5` 행을 v1.1.77에서 추가 완료. R006의 v2.1.267 DETAIL 노트(Opus 4.7·4.8·Fable 5의 `effort:` 프론트매터 무시 결함이 267에서 수정됐다는 기록)에는 #1720에서 v2.1.280의 후속 변경("Changed Opus 4.7, Opus 4.8 and Fable 5 to stop holding their launch-default effort over `/effort` in `-p` or the Agent SDK, a project, managed or `--settings` `effortLevel`, or a per-model level")을 반영한 후속 노트를 추가했습니다.
|
|
1396
1396
|
- 그 외 항목은 CC 플랫폼 신뢰성 개선이며 이 저장소 harness 변경은 불필요합니다.
|
|
1397
1397
|
|
|
1398
1398
|
---
|
package/templates/manifest.json
CHANGED