Compare commits

...

10 Commits

Author SHA1 Message Date
Drew Ritter
6e95f8dbdb Remove session scope language from SUBAGENT-STOP
Drop 'top-level sessions only' which may cause the model to treat
using-superpowers as a first-turn-only skill, breaking skill chaining
on follow-up turns.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-11 13:47:26 -08:00
Drew Ritter
5b3263d725 Soften SUBAGENT-STOP to only skip this skill, not suppress all behavior
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-11 12:58:30 -08:00
Drew Ritter
6023d5e74b Remove skill override directives from dispatch templates
Let the SUBAGENT-STOP gate in using-superpowers handle skill leakage
instead of per-template directives. This avoids blocking non-superpowers
skills that users may want subagents to use.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-11 12:55:24 -08:00
Drew Ritter
a6f3dd1614 Move Codex tool mapping to progressive disclosure reference file
Addresses Jesse's review feedback on PR #450:
- Move inline routing table from using-superpowers to references/codex-tools.md,
  leveraging Codex's native progressive disclosure for companion files
- Narrow SUBAGENT-STOP from "Do not invoke skills" to "Do not invoke
  superpowers skills" so subagents can still use non-superpowers skills

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-11 12:45:31 -08:00
Drew Ritter
a171953947 Sharpen subagent skill override directive in dispatch templates
Replace verbose scope explanations with a direct override statement
that explicitly claims priority over prior guidance (i.e. the
using-superpowers 1% rule injected by Codex's skill discovery).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-11 10:47:02 -08:00
Drew Ritter
c60991457a Document collab feature requirement for Codex subagent skills
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-11 10:47:02 -08:00
Drew Ritter
c8ede5e5c5 Prevent Codex subagent skill leakage via gate check and dispatch routing table
Codex subagents inherit filesystem access and can discover superpowers skills
via native discovery. Without guidance, they activate the 1% rule and invoke
full skill workflows instead of executing their assigned task.

- Add SUBAGENT-STOP gate check above the 1% rule in using-superpowers
- Add Codex dispatch routing table (spawn_agent/wait/close_agent)
- Add scope directives to all 4 subagent dispatch templates

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-11 10:47:02 -08:00
Jesse Vincent
4d54210a3c fix: restore polyglot wrapper to fix Windows hook window spawning
Claude Code spawns hook commands with shell:true + windowsHide:true,
but on Windows the execution chain cmd.exe -> bash.exe causes Git
Bash (MSYS2) to allocate its own console window, bypassing the hide
flag. This creates visible terminal windows that steal focus on every
SessionStart event (startup, resume, clear, compact).

The fix:
- Rename session-start.sh to session-start (no extension) so Claude
  Code's .sh auto-detection regex doesn't fire and prepend "bash"
- Restore run-hook.cmd polyglot wrapper to control bash invocation
  on Windows (tries known Git Bash paths, then PATH, then exits
  silently if no bash found)
- On Unix, the polyglot's shell portion runs the script directly

This avoids Claude Code's broken .sh auto-prepend, gives us control
over how bash is invoked on Windows, and gracefully handles missing
bash instead of erroring.

Addresses: #440, #414, #354, #417, #293
Upstream: anthropics/claude-code#14828
2026-02-10 18:34:58 -08:00
Jesse Vincent
8abf9d0b02 Merge pull request #443 from obra/wip/project-decomposition
Add project-level scope assessment to brainstorming pipeline
2026-02-10 10:42:20 -08:00
Jesse Vincent
a1a1ae5519 Merge pull request #441 from obra/wip/architecture-guidance
Add architecture guidance and capability-aware escalation
2026-02-10 10:40:57 -08:00
7 changed files with 73 additions and 30 deletions

1
.gitattributes vendored
View File

@@ -1,5 +1,6 @@
# Ensure shell scripts always have LF line endings
*.sh text eol=lf
hooks/session-start text eol=lf
# Ensure the polyglot wrapper keeps LF (it's parsed by both cmd and bash)
*.cmd text eol=lf

View File

@@ -32,6 +32,12 @@ Fetch and follow instructions from https://raw.githubusercontent.com/obra/superp
3. Restart Codex.
4. **For subagent skills** (optional): Skills like `dispatching-parallel-agents` and `subagent-driven-development` require Codex's collab feature. Add to your Codex config:
```toml
[features]
collab = true
```
### Windows
Use a junction instead of a symlink (works without Developer Mode):

View File

@@ -6,7 +6,7 @@
"hooks": [
{
"type": "command",
"command": "${CLAUDE_PLUGIN_ROOT}/hooks/session-start.sh",
"command": "${CLAUDE_PLUGIN_ROOT}/hooks/run-hook.cmd session-start",
"async": true
}
]

View File

@@ -1,43 +1,46 @@
: << 'CMDBLOCK'
@echo off
REM ============================================================================
REM DEPRECATED: This polyglot wrapper is no longer used as of Claude Code 2.1.x
REM ============================================================================
REM Cross-platform polyglot wrapper for hook scripts.
REM On Windows: cmd.exe runs the batch portion, which finds and calls bash.
REM On Unix: the shell interprets this as a script (: is a no-op in bash).
REM
REM Claude Code 2.1.x changed the Windows execution model for hooks:
REM Hook scripts use extensionless filenames (e.g. "session-start" not
REM "session-start.sh") so Claude Code's Windows auto-detection -- which
REM prepends "bash" to any command containing .sh -- doesn't interfere.
REM
REM Before (2.0.x): Hooks ran with shell:true, using the system default shell.
REM This wrapper provided cross-platform compatibility by
REM being both a valid .cmd file (Windows) and bash script.
REM
REM After (2.1.x): Claude Code now auto-detects .sh files in hook commands
REM and prepends "bash " on Windows. This broke the wrapper
REM because the command:
REM "run-hook.cmd" session-start.sh
REM became:
REM bash "run-hook.cmd" session-start.sh
REM ...and bash cannot execute a .cmd file.
REM
REM The fix: hooks.json now calls session-start.sh directly. Claude Code 2.1.x
REM handles the bash invocation automatically on Windows.
REM
REM This file is kept for reference and potential backward compatibility.
REM ============================================================================
REM
REM Original purpose: Polyglot wrapper to run .sh scripts cross-platform
REM Usage: run-hook.cmd <script-name> [args...]
REM The script should be in the same directory as this wrapper
if "%~1"=="" (
echo run-hook.cmd: missing script name >&2
exit /b 1
)
"C:\Program Files\Git\bin\bash.exe" -l "%~dp0%~1" %2 %3 %4 %5 %6 %7 %8 %9
exit /b
set "HOOK_DIR=%~dp0"
REM Try Git for Windows bash in standard locations
if exist "C:\Program Files\Git\bin\bash.exe" (
"C:\Program Files\Git\bin\bash.exe" "%HOOK_DIR%%~1" %2 %3 %4 %5 %6 %7 %8 %9
exit /b %ERRORLEVEL%
)
if exist "C:\Program Files (x86)\Git\bin\bash.exe" (
"C:\Program Files (x86)\Git\bin\bash.exe" "%HOOK_DIR%%~1" %2 %3 %4 %5 %6 %7 %8 %9
exit /b %ERRORLEVEL%
)
REM Try bash on PATH (e.g. user-installed Git Bash, MSYS2, Cygwin)
where bash >nul 2>nul
if %ERRORLEVEL% equ 0 (
bash "%HOOK_DIR%%~1" %2 %3 %4 %5 %6 %7 %8 %9
exit /b %ERRORLEVEL%
)
REM No bash found - exit silently rather than error
REM (plugin still works, just without SessionStart context injection)
exit /b 0
CMDBLOCK
# Unix shell runs from here
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
# Unix: run the named script directly
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]:-$0}")" && pwd)"
SCRIPT_NAME="$1"
shift
"${SCRIPT_DIR}/${SCRIPT_NAME}" "$@"
exec bash "${SCRIPT_DIR}/${SCRIPT_NAME}" "$@"

View File

@@ -3,6 +3,10 @@ name: using-superpowers
description: Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions
---
<SUBAGENT-STOP>
If you were dispatched as a subagent to execute a specific task, skip this skill.
</SUBAGENT-STOP>
<EXTREMELY-IMPORTANT>
If you think there is even a 1% chance a skill might apply to what you are doing, you ABSOLUTELY MUST invoke the skill.
@@ -27,6 +31,10 @@ If CLAUDE.md says "don't use TDD" and a skill says "always use TDD," follow CLAU
**In other environments:** Check your platform's documentation for how skills are loaded.
## Platform Adaptation
Skills use Claude Code tool names. Non-CC platforms: see `references/codex-tools.md` for tool equivalents.
# Using Skills
## The Rule

View File

@@ -0,0 +1,25 @@
# Codex Tool Mapping
Skills use Claude Code tool names. When you encounter these in a skill, use your platform equivalent:
| Skill references | Codex equivalent |
|-----------------|------------------|
| `Task` tool (dispatch subagent) | `spawn_agent` |
| Multiple `Task` calls (parallel) | Multiple `spawn_agent` calls |
| Task returns result | `wait` |
| Task completes automatically | `close_agent` to free slot |
| `TodoWrite` (task tracking) | `update_plan` |
| `Skill` tool (invoke a skill) | Skills load natively — just follow the instructions |
| `Read`, `Write`, `Edit` (files) | Use your native file tools |
| `Bash` (run commands) | Use your native shell tools |
## Subagent dispatch requires collab
Add to your Codex config (`~/.codex/config.toml`):
```toml
[features]
collab = true
```
This enables `spawn_agent`, `wait`, and `close_agent` for skills like `dispatching-parallel-agents` and `subagent-driven-development`.