oh-my-customcodex 0.5.17 → 0.5.19

package/dist/cli/index.js

@@ -3091,7 +3091,7 @@ var init_package = __esm(() => {
     workspaces: [
       "packages/*"
     ],
-    version: "0.5.17",
+    version: "0.5.19",
     requiresCC: ">=2.1.121",
     claudeCode: {
       minimumVersion: "2.1.121",

package/dist/index.js

@@ -2316,7 +2316,7 @@ var package_default = {
   workspaces: [
     "packages/*"
   ],
-  version: "0.5.17",
+  version: "0.5.19",
   requiresCC: ">=2.1.121",
   claudeCode: {
     minimumVersion: "2.1.121",

package/package.json

@@ -3,7 +3,7 @@
   "workspaces": [
     "packages/*"
   ],
-  "version": "0.5.17",
+  "version": "0.5.19",
   "requiresCC": ">=2.1.121",
   "claudeCode": {
     "minimumVersion": "2.1.121",

package/templates/.claude/rules/MUST-completion-verification.md

@@ -85,8 +85,10 @@ Before invoking or registering generated workflow code, run a Tier-1 sanity pass
 
 1. Search for unresolved placeholders, malformed identifiers, and known dead-line patterns.
 2. Assemble the exact script body before execution; do not concatenate facts or guardrails after the call starts.
-3. Run the language parser or the narrowest no-side-effect syntax check available.
-4. If syntax validation is unavailable, read the generated body back and perform a focused self-review before launch.
+3. For compound shell verification or release gates, start the script with `set -euo pipefail`; `set -o pipefail` alone does not stop later sequential gates after an earlier command fails.
+4. For zsh/bash snippets, avoid reserved or special variable names such as `status`, `path`, and `argv`; use `run_status`, `cmd_path`, `args`, or similarly specific names.
+5. Run the language parser or the narrowest no-side-effect syntax check available.
+6. If syntax validation is unavailable, read the generated body back and perform a focused self-review before launch.
 
 This check is mandatory for workflow scripts that will run automation, mutate GitHub state, or gate a release.
 
@@ -162,6 +164,10 @@ When a user sends a new instruction while work is in progress, completion status
 3. If the new message adds a requirement, add it to the completion contract before closing.
 4. If no conflict exists, continue but explicitly preserve the new requirement in the next verification pass.
 
+### Tool-Call Payload Completeness
+
+도구 호출의 required 파라미터는 invoke 전에 확인한다(완료 선언 후가 아니라 호출 시점의 전제조건). announce(prefix)만 출력하고 payload의 required 필드를 누락하는 패턴은 R008 "Required-Parameter Completeness Check"가 canonical owner다. Reference: #1487 / upstream #1324.
+
 ## Completion Contract Format — [Contract] + [Done] with criterion/evidence pairs. See template via Read tool.
 
 <!-- DETAIL: Completion Contract Format

package/templates/.claude/rules/MUST-parallel-execution.md

@@ -34,6 +34,19 @@ Before writing/editing multiple files:
 5. Running agent stalled (2x+ duration)? → Spawn independent follow-up tasks immediately
 6. Parallel dispatch announced? → put all announced tool calls in the same message
 
+### LLM Batch Output Token Budget
+
+For N-item structured LLM batches, pre-compute output tokens and chunk variable-size lists (default ≤40 items); raising `max_tokens` alone is not enough. Reference: #1486 / upstream #1321 and #1320.
+
+<!-- DETAIL: LLM Batch Output Token Budget
+The giant-prompt heuristic governs input tokens. The symmetric output-side rule: when a single LLM call processes N items (scoring/classifying/extracting) and must emit structured output (for example JSON) per item, pre-compute the output budget = N × per-item output tokens before the call. Exceeding `max_tokens` truncates the response mid-structure, so the call can appear to succeed while JSON parsing fails.
+
+| Anti-pattern | Required |
+|--------------|----------|
+| Single batch call over a variable-size list with a fixed small `max_tokens` | Chunk into bounded batches (default ≤40 items), constrain per-item output length, and set `max_tokens` to fit one chunk |
+| Raising `max_tokens` alone | Insufficient — it only defers the failure as the list grows. Chunking is the invariant fix. |
+-->
+
 ### Common Violations to Avoid
 
 <!-- DETAIL: Common Violations Short Examples

package/templates/.claude/rules/MUST-tool-identification.md

@@ -50,6 +50,18 @@ Incorrect parallel: tool_call(url1), tool_call(url2), tool_call(cmd) — no iden
 Correct parallel: list ALL [agent][model] → Tool/Fetching/Running lines FIRST, then all tool_calls
 -->
 
+
+### Required-Parameter Completeness Check
+
+R008 prefix(announce)와 실제 도구 호출은 분리된 단계다. prefix를 출력한 뒤 호출 payload에서 도구 스키마상 required 파라미터를 누락하면 호출이 실패하거나 빈 동작이 된다. 호출 직전, prefix 존재뿐 아니라 required 파라미터가 모두 채워졌는지 확인한다.
+
+| Anti-pattern | Required |
+|--------------|----------|
+| `[agent][model] → Tool: AskUserQuestion` prefix만 출력하고 `questions` 파라미터 없이/빈 배열로 호출 | prefix + `questions` 배열(최소 1개) 모두 채워 호출 |
+| announce 후 payload의 required 필드 누락 (announce-payload separation gap) | announce와 동일 메시지에서 required 필드 완비 호출 |
+
+Cross-reference: R020 (action-completeness precondition — invoke 전에 required 파라미터 확인). Reference issue: #1487 / upstream #1324 (AskUserQuestion `questions` 누락 재발 방지).
+
 ## Models
 
 | Model | Use |
@@ -100,6 +112,31 @@ matches the spawn announcement:
   [2] lang-python-expert:sonnet → Python code review
 ```
 
+## Tier-3 Interaction Tool Prefix (MANDATORY)
+
+R008 "every tool call" applies to Tier-3 interaction tools too — not only file/exec tools. Applying the `[agent][model] → Tool:` prefix to Bash/Read/Agent while omitting it on `AskUserQuestion`, `TodoWrite`, `EnterPlanMode`, `ExitPlanMode`, `request_user_input`, or equivalent structured-question tools is a violation.
+
+| Tool | R008 prefix required? |
+|------|----------------------|
+| AskUserQuestion / `request_user_input` / structured question | YES — `[agent][model] → Tool: AskUserQuestion` or equivalent before the call |
+| TodoWrite / `update_plan` | YES |
+| EnterPlanMode / ExitPlanMode | YES |
+| Skill | NO separate R008 prefix — identified via the R007 integrated header instead |
+
+Skill invocation is the one exception: it is identified through the R007 integrated identification block (`┌─ Agent: {agent} → {skill-name}`), not a standalone R008 tool prefix.
+
+Reference issue: #1486 / upstream #1321 (AskUserQuestion prefix omission); complements #1487 required-payload completeness.
+
+## Multi-Turn Self-Check
+
+도구 호출 전 매번 확인한다:
+
+1. 이 호출 위에 `[agent-name][model] → Tool: <tool-name>` 라인이 있는가?
+2. agent-name과 model이 현재 컨텍스트와 일치하는가?
+3. 이 호출에 도구 스키마상 required 파라미터가 모두 채워져 있는가? 예: AskUserQuestion/request_user_input 계열은 `questions` 배열이 비어 있지 않아야 한다. prefix(announce)만 출력하고 실제 호출 payload의 required 필드를 누락하면 안 된다.
+
+체크 실패 시 즉시 prefix/필수 파라미터를 보완한 후 호출.
+
 ## Example
 
 ```

package/templates/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.5.17",
+  "version": "0.5.19",
   "requiresCC": ">=2.1.121",
   "claudeCode": {
     "minimumVersion": "2.1.121",

package/templates/workflows/auto-dev.yaml

@@ -126,6 +126,11 @@ steps:
       - If a scoped issue is added or materially changed after verify-build starts, reset this phase and re-run the full verify-build gate for the new aggregate scope.
       - Do not proceed to release from a verify-build result collected before the latest scoped implementation change.
 
+      Shell execution discipline:
+      - Run compound verification commands under `set -euo pipefail` so lint, typecheck, tests, and build halt on the first failed gate.
+      - Do not rely on `set -o pipefail` alone for sequential verification chains.
+      - Avoid reserved shell variable names in polling or status snippets (`status`, `path`, `argv`); use `run_status`, `cmd_path`, or `args` instead.
+
       For Node/Bun projects:
       1. Run `bun install` and halt on lockfile drift.
       2. Run `bun run lint` when available.