oh-my-customcode 2.0.7 → 2.0.9

package/templates/.claude/skills/goal/SKILL.md

@@ -1,21 +1,21 @@
 ---
 name: omcustom:goal
-description: Disciplined goal-to-execution workflow for any user task. Parses objective, asks only for materially missing requirements via ambiguity-gate, inspects repo via idea, plans via sdd-dev or deep-plan, executes safely under project conventions, verifies completion per R020, and reports changed files with evidence. Use when user invokes /goal <task>.
+description: Disciplined goal-to-execution workflow for any user task. Parses objective, asks only for materially missing requirements via ambiguity-gate, inspects repo via idea, plans via sdd-dev or deep-plan, executes safely under project conventions, verifies completion per R020, and reports changed files with evidence. Use when user invokes /omcustom:goal <task>.
 version: 1.0.0
 scope: core
 ---
 
-# /goal — Disciplined Goal-to-Execution Workflow
+# /omcustom:goal — Disciplined Goal-to-Execution Workflow
 
 Apply project discipline (planning, verification, cleanup, review) to any user-provided task. Acts as a thin orchestrator that routes to existing oh-my-customcode skills rather than duplicating logic.
 
 ## Usage
 
 ```
-/goal <task description>
+/omcustom:goal <task description>
 ```
 
-Example: `/goal Add a /weekly-summary command that aggregates last 7 days of git activity`
+Example: `/omcustom:goal Add a /weekly-summary command that aggregates last 7 days of git activity`
 
 ## Workflow
 
@@ -73,7 +73,7 @@ This skill is intentionally a **thin wrapper / orchestrator**. It does NOT dupli
 - Stable entry point that routes to whichever underlying skills are best at the moment
 - Documented contract that callers can rely on regardless of internal skill evolution
 
-If a future skill replaces `idea` or `sdd-dev`, only the integration table updates — `/goal` remains stable.
+If a future skill replaces `idea` or `sdd-dev`, only the integration table updates — `/omcustom:goal` remains stable.
 
 ## Cross-References
 

package/templates/.claude/skills/model-escalation/SKILL.md

@@ -19,33 +19,37 @@ haiku → sonnet → opus
 
 | Condition | Action |
 |-----------|--------|
-| 2+ failures with same model for same agent type | Advise escalation |
-| 3+ consecutive failures across any agent type | Advise global escalation |
-| Sustained success after escalation | Advise de-escalation |
+| 2+ recorded failures for the same agent type (any model; evaluated only for haiku/sonnet spawns) | Advise escalation |
+| 3+ consecutive failures across any agent type | Advise escalation (per spawn, not global; only when the spawn's `model` parameter is haiku or sonnet) |
+| Sustained success after escalation | Advise de-escalation (opt-in: `OMCUSTOM_MODEL_DEESCALATION=on`) |
 
 ## Thresholds
 
-| Parameter | Default | Description |
-|-----------|---------|-------------|
-| `failure_threshold` | 2 | Failures before escalation advisory |
-| `consecutive_threshold` | 3 | Consecutive failures for global advisory |
-| `cooldown_tasks` | 5 | Successes before de-escalation advisory |
+Hard-coded in `model-escalation-advisor.sh` (not configurable via frontmatter or settings):
+
+| Constant | Value | Description |
+|----------|-------|-------------|
+| `FAILURE_THRESHOLD` | 2 | Failures before escalation advisory |
+| `CONSECUTIVE_THRESHOLD` | 3 | Consecutive failures for the per-spawn advisory |
+| `COOLDOWN` | 5 | Successes before de-escalation advisory (opt-in only) |
+
+Each distinct advisory is shown at most once per session (atomic per-key marker dir `/tmp/.claude-model-advisory-shown-${PPID}.d/`). De-escalation is off by default because `task-outcome-recorder.sh` records every subagent completion as success, so consecutive successes carry no signal.
 
 ## Cost Guard
 
 - Advisory includes estimated cost multiplier
-- De-escalation suggested after sustained success at higher tier
+- De-escalation (opt-in via `OMCUSTOM_MODEL_DEESCALATION=on`) suggested after sustained success at higher tier
 - Cost tracked per session via PPID-scoped temp file
 
 ## Architecture
 
 ```
-PostToolUse (Task) → task-outcome-recorder.sh
-  Records: agent_type, model, success/failure, error_summary
+SubagentStop → task-outcome-recorder.sh (the only wiring; not PostToolUse)
+  Records: agent_type, model ("inherit"), outcome (always success; no failure signal, so the escalation advisory is currently unreachable)
   Storage: /tmp/.claude-task-outcomes-$PPID (JSON lines, max 50)
 
 PreToolUse (Task) → model-escalation-advisor.sh
-  Reads outcomes → counts failures → advises escalation via stderr
+  Reads outcomes → counts failures → advises escalation via additionalContext
   Advisory only — never blocks, never modifies tool input
 ```
 

package/templates/.claude/skills/omcustom-loop/SKILL.md

@@ -15,19 +15,19 @@ When background subagents complete, `SubagentStop` command hooks record outcomes
 ## How It Works
 
 1. When a background subagent completes, Claude Code fires the `SubagentStop` event
-2. The command hooks run: `task-outcome-recorder.sh` (records an outcome entry, but SubagentStop exposes no failure signal, so entries are always outcome=success; see the "Outcome caveat" comment in the script), `stall-detection-advisor.sh` (stderr advisory on stalled parallel agents; stderr on exit 0 is visible to humans in transcript/debug, not fed to the model), `auto-continue-guard.sh` (counts consecutive completions), `r007-r008-drift-advisor.sh` (R007/R008 drift advisory), `subagent-failure-advisor.sh` (reads `.tool_output.is_error`, which is absent from the SubagentStop payload, so its warning can never trigger on this path until #1656 G (closed unfixed; tracked in #1775); the "Outcome caveat" comment in `task-outcome-recorder.sh` and the `is_error=` line of the advisor)
-3. None of them continues the orchestrator's workflow; the orchestrator decides the next workflow step itself from the returned results. Exception: when `r007-r008-drift-advisor.sh` emits an advisory on SubagentStop, CC delivers the `additionalContext` to the stopping subagent, which continues its turn to respond (CHANGELOG 2.1.163).
+2. The command hooks run: `task-outcome-recorder.sh` (records an outcome entry, but SubagentStop exposes no failure signal, so entries are always outcome=success (hooks doc lists SubagentStop input fields as `stop_hook_active`, `agent_id`, `agent_type`, `agent_transcript_path`, `last_assistant_message`, `background_tasks`, `session_crons` plus common fields; measured payload keys: https://github.com/baekenough/oh-my-customcode/issues/1775#issuecomment-6404861043); see the "Outcome caveat" comment in the script), `stall-detection-advisor.sh` (stderr advisory on stalled parallel agents; stderr on exit 0 is not fed to the model (the doc says only: "To read it yourself … enable debug logging"); official hooks doc https://code.claude.com/docs/en/hooks.md: "Claude never sees stderr from a hook that exits 0."), `auto-continue-guard.sh` (counts consecutive completions), `r007-r008-drift-advisor.sh` (R007/R008 drift advisory); failure advisories for Agent/Task calls come from `PostToolUseFailure` (`subagent-failure-advisor.sh`) instead, not from SubagentStop
+3. None of them continues the orchestrator's workflow; the orchestrator decides the next workflow step itself from the returned results. Exception: when `r007-r008-drift-advisor.sh` emits an advisory on SubagentStop, CC delivers the `additionalContext` to the stopping subagent, which continues its turn to respond (CHANGELOG 2.1.163); hooks doc: SubagentStop hooks support `hookSpecificOutput.additionalContext` "for non-error feedback that keeps the subagent running", and "To inject context into the parent session after a subagent returns, use a [`PostToolUse`](#posttooluse) hook on the `Agent` tool instead."
 4. If no steps remain, it reports results and waits for user input
 
 ## Consecutive-Completion Advisory
 
-- **Consecutive-completion counter**: `auto-continue-guard.sh` counts SubagentStop events (counter resets after 60s idle); from the 4th completion onward within the window (count >= 4) it prints a stderr line "limit (3) reached. Pausing." (human-visible only, not a model-visible signal; among these five SubagentStop hooks only `r007-r008-drift-advisor.sh` uses `additionalContext`) and its script contains no `exit 2` statement (its only explicit exit is `exit 0`); it echoes the input payload to stdout, and the CC 2.1.310 SubagentStop payload has no top-level `decision`/`continue` key (CC 2.1.310 binary SubagentStop payload builder, inspected during the v2.0.6/v2.0.7 reviews), so it does not block
+- **Consecutive-completion counter**: `auto-continue-guard.sh` counts SubagentStop events (counter resets after more than 60 s idle since the last completion, so the window chains); from the 4th completion onward within the window (count >= 4) it prints a stderr line "[ConsecutiveCompletions] ADVISORY: 4+ consecutive subagent completions, each within 60s of the previous (advisory only; nothing is paused)." (human-visible only, not a model-visible signal; among these four SubagentStop hooks only `r007-r008-drift-advisor.sh` uses `additionalContext`) and its script contains no `exit 2` statement (its only explicit exit is `exit 0`); it echoes the input payload to stdout, and the CC 2.1.310 SubagentStop payload has no top-level `decision`/`continue` key (CC 2.1.310 binary SubagentStop payload builder, inspected during the v2.0.6/v2.0.7 reviews), so it does not block
 - **stuck-detector integration**: If the same action repeats 3+ times, the independent PostToolUse stuck-detector hook intervenes
 - **cost-cap-advisor**: Cost monitoring is independent of these hooks
 
 ## Configuration
 
-The hooks are configured in `.claude/hooks/hooks.json` under `SubagentStop`. All five hooks are `type: command` and advisory (none contains an `exit 2` statement or emits a `decision`/`continue` field outside comments, so none blocks; the grep pattern covered `exit 2`, `"decision"`, `decision:`, `"continue"` and `continue:` and hit only comments in `r007-r008-drift-advisor.sh`; the `continue:` form added no hits. Four of the five hooks echo the CC payload to stdout, which carries no top-level `decision`/`continue` key in CC 2.1.310. The drift advisor's `additionalContext` makes the stopping subagent continue its own turn; that continuation is not a block).
+The hooks are configured in `.claude/hooks/hooks.json` under `SubagentStop`. All four hooks are `type: command` and advisory (none contains an `exit 2` statement or emits a `decision`/`continue` field outside comments, so none blocks; the grep pattern covered `exit 2`, `"decision"`, `decision:`, `"continue"` and `continue:` and hit only comments in `r007-r008-drift-advisor.sh`; the `continue:` form added no hits. Three of the four hooks (`task-outcome-recorder.sh`, `stall-detection-advisor.sh`, `auto-continue-guard.sh`; measured by `git grep -n` of `printf '%s\n' "$input"` at task-outcome-recorder.sh:218, stall-detection-advisor.sh:31/:44/:124, auto-continue-guard.sh:32; `r007-r008-drift-advisor.sh` instead prints the `additionalContext` JSON on stdout, :456-:457; its advisory text also goes to stderr at :451, which the model does not see) echo the CC payload to stdout, which carries no top-level `decision`/`continue` key in CC 2.1.310. The drift advisor's `additionalContext` makes the stopping subagent continue its own turn; that continuation is not a block).
 
 ## Limitations
 

package/templates/.claude/skills/pipeline/workflows/auto-dev.yaml

@@ -928,6 +928,19 @@ steps:
         각 변수에 최소 한 값 변경 대조 조건을 두게 하십시오. 측정 writer는 조건·결과 표를 제출하고
         일반화 문장에 측정 조건을 함께 적으십시오. dispatcher는 대조가 없는 함의를 가설로
         제한하며 표 없는 runtime 일반화를 완료로 받지 마십시오(#1833 P5).
+      - Line-anchor drift sweep: a delegation that adds or removes lines in any file MUST, as a
+        completion condition, run `git grep -n -E '<escaped basename>:[0-9]'` across the
+        repository (the BASENAME, not the path — basename-only anchors such as
+        `task-outcome-recorder.sh:218` carry no directory) and also check every sentence that
+        mentions the file for bare continuation anchors (`:[0-9]+` after a path mention, e.g.
+        `stall-detection-advisor.sh:44/:124` or `:451-:453`), then either re-derive each anchor
+        whose target text moved or replace it with a content anchor (function name / unique
+        string). Anchors into generated files (e.g. `.claude/settings.json`, regenerated from
+        hooks.json) MUST be content anchors, not line numbers. The completion report lists each
+        anchor (file:line, old target, new line).
+        (줄 수를 바꾸는 위임의 완료 조건에 그 파일을 가리키는 `file:line` 앵커 — basename 전용·
+        연속 `:N` 형태 포함 — 의 전수 grep과 재도출 또는 내용 앵커 전환을 넣어야 하며, 생성 파일
+        앵커는 줄 번호가 아닌 내용 앵커여야 합니다) (#1884 item 8).
       - No state-changing command (`gh issue create/edit/close`, `gh pr create/edit/close/merge`,
         `git push`, label changes) — including the orchestrator's own direct commands, not only
         delegated ones — MAY use a `||` fallback that prints success-looking output; the observed

package/templates/guides/claude-code/04-agent-skills.md

@@ -55,7 +55,7 @@ description: Brief description of what this Skill does and when to use it
 
 ### 위치 인자 `$N` 치환 (#1879)
 
-Claude Code 2.1.310에서 스킬을 인자 3개(`alpha beta gamma`)와 함께 호출해 관측한 결과, `SKILL.md` 안의 인자 범위 안의 인덱스(이번 관측 `$0`~`$2`)는 bash 펜스 코드 블록, 큰따옴표, 함수 본문, 산문에서 인자 값으로 치환되었습니다(`"$1"` → `"beta"`). `${1}` 표기는 치환되지 않았고, 두 번째 관측(스킬 도구로 omcustom-feedback을 인자 `alpha beta gamma`와 함께 로드, CC 2.1.310)에서도 `"${1}"`, `[ -f "${1}" ]`, `kind="${2}"`가 리터럴로 남았으므로, 스킬의 셸 코드에서는 위치 매개변수를 `${1}`로 쓰십시오. 범위를 벗어난 인덱스는 산문의 `$5`만 관측했고 리터럴로 남았으며, 인자 없는 호출도 치환되지 않았고, `$10`도 3개 인자 호출에서는 리터럴로 남았습니다(한 자리 숫자만 인식하는지는 확정하지 못했습니다 [가설]). 인자가 있는 호출에서는 산문의 `$5` 같은 금액 표기도 같은 규칙으로 치환될 수 있으므로, 금액은 `USD 5`처럼 쓰는 것을 권장합니다.
+Claude Code 2.1.310에서 스킬을 인자 3개(`alpha beta gamma`)와 함께 호출해 관측한 결과, `SKILL.md` 안의 인자 범위 안의 인덱스(이번 관측 `$0`~`$2`)는 bash 펜스 코드 블록, 큰따옴표, 함수 본문, 산문에서 인자 값으로 치환되었습니다(`"$1"` → `"beta"`). `${1}` 표기는 치환되지 않았고, 두 번째 관측(스킬 도구로 omcustom-feedback을 인자 `alpha beta gamma`와 함께 로드, CC 2.1.310)에서도 `"${1}"`, `[ -f "${1}" ]`, `kind="${2}"`가 리터럴로 남았으므로, 스킬의 셸 코드에서는 위치 매개변수를 `${1}`로 쓰십시오. 범위 밖이 확실한 인덱스는 산문의 `$5`만 관측했고 리터럴로 남았으며, 인자 없는 호출도 치환되지 않았고, `$10`도 3개 인자 호출에서는 리터럴로 남았습니다(한 자리 숫자만 인식하는지는 확정하지 못했습니다 [가설]). 인자가 있는 호출에서는, 인자 6개 이상이면 산문의 `$5` 같은 금액 표기도 같은 규칙으로 치환될 수 있으므로, 금액은 `USD 5`처럼 쓰는 것을 권장합니다.
 
 ## 사용 가능한 Skills