Delete v4-era planning/internal docs, stale translations, the broken npm
wrapper plus redundant install scripts (install.sh/setup.py/MANIFEST.in),
and the v4 docs tree. Rewrite CONTRIBUTING for the v5 eval gate, fix dead
links in SECURITY/docs, English-only README. pipx/uv + superclaude CLI is
the single install path. Remove translation/cross-repo-sync CI.
- manifest/metadata.json: plugin/marketplace metadata consumed by
scripts/build_superclaude_plugin.py; carries no real version.
- manifest/plugin.template.json, marketplace.template.json: rendered
into dist .claude-plugin/{plugin,marketplace}.json. Marketplace
format per code.claude.com/docs/en/plugin-marketplaces (name, owner,
plugins[]).
- build script: VERSION file (PEP 440, 5.0.0a1) stays the single
source of truth; new pep440_to_semver() renders it as 5.0.0-alpha.1
to match the checked-in plugin.json. Also copy .mcp.json into dist
since the rendered plugin.json references it.
- .gitignore: anchor MANIFEST to repo root; the unanchored pattern
case-insensitively swallowed plugins/superclaude/manifest/ on macOS.
dist output verified: 4 skills, 1 agent, hooks.json + executable
scripts, rendered manifests; no commands/ or modes/.
- session-restore.sh (SessionStart, command): emits TASK.md head, git
branch/status/log, and probes mindbase on :18002/:18003 with 1s
timeout; every probe degrades silently, always exits 0. Replaces
session-init.sh (its git summary folded in; banner text dropped).
- confidence-gate (PreToolUse Write|Edit, prompt): enforces
confidence-check skill criteria before non-trivial new
implementations; trivial edits and test fixes exempt.
- session-summary (Stop, prompt): 2-3 bullet outcome summary, stored
via mindbase_store when available.
- reflexion-trigger (Stop, prompt): applies pm-reflexion
failure-reflexion when a session hit real errors.
- tab-title.sh: ported from airis-tab-title.sh, opt-in via
SUPERCLAUDE_TAB_TITLE=1, silent no-op otherwise; wired to
SessionStart/UserPromptSubmit/PreToolUse/PostToolUse/Notification/
Stop to keep the idle/running/waiting/stop state machine working.
- statusline.sh: genericized statusline asset (not wired into
hooks.json; users reference it from settings).
Hook schema verified against code.claude.com/docs/en/hooks:
type=command and type=prompt confirmed; timeout is in seconds.
- 'superclaude install' now installs all skills to ~/.claude/skills/
and agents to ~/.claude/agents/; new --minimal flag installs only
the confidence-check skill with no agents
- 'superclaude update' force-reinstalls the same set
- Rename install_commands.py -> install_assets.py (agents only);
remove dead install_commands/list_*_commands functions
- install_skill: dual source resolution (package dir + plugins/),
new install_all_skills helper
- doctor: recognize SKILL.md skill layout (legacy implementation.md
still accepted)
- install.sh / Makefile: drop slash-command install/sync steps
- Update unit and e2e tests to the new install behavior
SuperClaude v5 slims down to an eval-gated toolkit:
- Delete all 30 slash commands and 7 behavioral modes (both
plugins/superclaude/ and src/superclaude/ mirrors)
- Delete all 20 persona agents; add a single explore-haiku agent
- Keep 4 skills: confidence-check, spec-panel, socratic, pm-reflexion
(mirrored into src/superclaude/skills/ for wheel installs)
- Drop brainstorm, deep-research, pm, token-efficiency, troubleshoot skills
Machine-scored A/B of candidate components vs native Claude Code
(claude -p baseline vs --plugin-dir variants/<comp>). Pre-registered
survive thresholds in preregister.yaml decide what ships in v5;
parallel.py is already a confirmed cut.
Default ReflexionPattern() no longer writes solutions_learned.jsonl or
docs/mistakes/*.md into the working directory. Persistence now requires
either an explicit memory_dir or the SUPERCLAUDE_REFLEXION_OUTPUT_DIR
env var. Stops the pytest plugin from polluting user repos on every
failed test. docs/mistakes/ is now gitignored.
The new test_mcp_list test (added in this PR) exposed a latent bug:
check_mcp_server_installed() spawns 'claude mcp list' but only caught
TimeoutExpired/SubprocessError, not FileNotFoundError. On machines
without the claude CLI (e.g. CI) the call raised FileNotFoundError and
crashed 'superclaude mcp --list'. Add FileNotFoundError to the except,
matching the pattern already used by check_docker_available() and the
prerequisite checks in this file.
Also apply ruff format to files this PR introduced/modified:
confidence.py, test_cli_e2e.py, test_execution_init.py.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
These violations exist on upstream master, unrelated to the confidence
work in this PR, but the lint CI job scans all of src/ tests/ so they
block this PR from going green. Fixing them here.
- remove unused imports (F401): test_parallel.py, test_execution_engine.py
- apply ruff format: install_commands.py, execution/__init__.py,
reflexion.py, test_reflection.py, test_self_correction.py
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Complete the confidence-check skill's TypeScript implementation (4 of its 5
checks were placeholder stubs that only read context flags) and enrich the
Python ConfidenceChecker with real tech-stack detection, architecture
anti-pattern warnings, and multi-language duplicate search.
Logic is ported from the kazukinakai fork. The dead airis-agent import layer
from the fork is intentionally dropped: the checks work with zero optional
dependencies, and each check still honors its explicit *_complete / *_verified
context-flag override for testing and pre-checked scenarios.
- confidence.ts: implement noDuplicates / architectureCompliant /
hasOssReference / rootCauseIdentified with project scanning, CLAUDE.md
tech-stack parsing, and regex-based uncertainty detection. Synced across all
four tracked skill copies (src CLI source, plugin source, repo-root, .claude).
- confidence.py: add _read_tech_stack and _check_architecture_anti_patterns;
multi-language (py/ts/js) + content-grep _search_codebase; richer root-cause
validation. Flat single-class structure preserved; accepts both upstream
(target_name/root_cause) and fork (feature_name/proposed_technology) keys.
- tests: add test_cli_main, test_doctor, test_execution_init, test_install_mcp,
test_install_skill, e2e/test_cli_e2e; expand test_confidence to a superset.
All ported tests pass against the existing upstream modules. fork-only
init/check CLI tests and circular full-suite meta-tests were excluded.
NOTE: the confidence-check skill is duplicated across four tracked locations;
a follow-up should collapse these to a single source.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
README Quality Check / Multi-language README Quality Assessment (push) Has been cancelled
Tests / Test on Python 3.10 (push) Has been cancelled
Tests / Test on Python 3.11 (push) Has been cancelled
Tests / Test on Python 3.12 (push) Has been cancelled
Tests / Lint and Format Check (push) Has been cancelled
Tests / Pytest Plugin Check (push) Has been cancelled
Tests / SuperClaude Doctor Check (push) Has been cancelled
Tests / Test Summary (push) Has been cancelled
* Add Windows installation guide for PowerShell users
Provides step-by-step instructions for installing SuperClaude on Windows
since install.sh only supports Linux/macOS and the existing docs assume
Unix tooling.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: sort imports to pass ruff I001 check
Pre-existing lint issue in execution/__init__.py that blocks CI.
---------
* fix: fill implementation gaps across core modules
- Replace ConfidenceChecker placeholder methods with real implementations
that search the codebase for duplicates, verify architecture docs exist,
check research references, and validate root cause specificity
- Fix intelligent_execute() error capture: collect actual errors from
failed tasks instead of hardcoded None, format tracebacks as strings,
and fix variable shadowing bug where loop var overwrote task parameter
- Implement ReflexionPattern mindbase integration via HTTP API with
graceful fallback when service is unavailable
- Fix .gitignore: remove duplicate entries, add explicit !-rules for
.claude/settings.json and .claude/skills/, remove Tests/ ignore
- Remove unnecessary sys.path hack in cli/main.py
- Fix FailureEntry.from_dict to not mutate input dict
- Add comprehensive execution module tests: 62 new tests covering
ParallelExecutor, ReflectionEngine, SelfCorrectionEngine, and the
intelligent_execute orchestrator (136 total, all passing)
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: include test-generated reflexion artifacts
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* fix: address 5 open GitHub issues (#536, #537, #531, #517, #534)
Security fixes:
- #536: Remove shell=True and user-controlled $SHELL from _run_command()
to prevent arbitrary code execution. Use direct list-based subprocess.run
without passing full os.environ to child processes.
- #537: Add SHA-256 integrity verification for downloaded docker-compose
and mcp-config files. Downloads are deleted on hash mismatch. Gateway
config supports pinned hashes via docker_compose_sha256/mcp_config_sha256.
Bug fixes:
- #531: Add agent file installation to `superclaude install` and `update`
commands. 20 agent markdown files are now copied to ~/.claude/agents/
alongside command installation.
- #517: Fix MCP env var flag from --env to -e for API key passthrough,
matching the Claude CLI's expected format.
Usability:
- #534: Replace Japanese trigger phrases and report labels in pm-agent.md
and pm.md (both src/ and plugins/) with English equivalents for
international accessibility.
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* docs: align documentation with Claude Code and fix version/count gaps
- Update CLAUDE.md project structure to include agents/ (20 agents),
modes/ (7 modes), commands/ (30 commands), skills/, hooks/, mcp/,
and core/ directories. Add Claude Code integration points section.
- Fix version references: 4.1.5 -> 4.2.0 in installation.md,
quick-start.md, and package.json (was 4.1.7)
- Fix feature counts across all docs:
- Commands: 21 -> 30
- Agents: 14/16 -> 20
- Modes: 6 -> 7
- MCP Servers: 6 -> 8
- Update README.md agent count from 16 to 20
- Add docs/user-guide/claude-code-integration.md explaining how
SuperClaude maps to Claude Code's native features (commands,
agents, hooks, skills, settings, MCP servers, pytest plugin)
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: update test-generated reflexion log
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* docs: comprehensive Claude Code gap analysis and integration guide
- Rewrite docs/user-guide/claude-code-integration.md with full feature
mapping: all 28 hook events, skills system with YAML frontmatter,
5 settings scopes, permission rules, plan mode, extended thinking,
agent teams, voice, desktop features, and session management.
Includes detailed gap table showing where SuperClaude under-uses
Claude Code capabilities (skills migration, hooks integration,
plan mode, settings profiles).
- Add Claude Code native features section to CLAUDE.md with extension
points we use vs should use more (hooks, skills, plan mode, settings)
- Add Claude Code integration gap analysis to KNOWLEDGE.md with
prioritized action items for skills migration, hooks leverage,
plan mode integration, and settings profiles
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: update test-generated reflexion log
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: bump version to 4.3.0
Bump version across all 15 files:
- VERSION, pyproject.toml, package.json
- src/superclaude/__init__.py, src/superclaude/__version__.py
- CLAUDE.md, PLANNING.md, TASK.md, CHANGELOG.md
- README.md, README-zh.md, README-ja.md, README-kr.md
- docs/getting-started/installation.md, quick-start.md
- docs/Development/pm-agent-integration.md
Also fixes __version__.py which was out of sync at 0.4.0.
Adds comprehensive CHANGELOG entry for v4.3.0.
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* i18n: replace all Japanese/Chinese text with English in source files
Replace CJK text with English across all non-translation files:
- src/superclaude/commands/pm.md: 38 Japanese strings in PDCA cycle,
error handling patterns, anti-patterns, document templates
- src/superclaude/agents/pm-agent.md: 20 Japanese strings in PDCA
phases, self-evaluation, documentation sections
- plugins/superclaude/: synced from src/ copies
- .github/workflows/readme-quality-check.yml: all Chinese comments,
table headers, report strings, and PR comment text
- .github/workflows/pull-sync-framework.yml: Japanese comment
- .github/PULL_REQUEST_TEMPLATE.md: complete rewrite from Japanese
Translation files (README-ja.md, docs/user-guide-jp/, etc.) are
intentionally kept in their respective languages.
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* feat: package SuperClaude as native Claude Code plugin
Create a fully-structured Claude Code plugin at plugins/superclaude/:
Plugin manifest (.claude-plugin/plugin.json):
- Name: superclaude, version 4.3.0
- Registers 30 commands, 20 agents, 6 skills, hooks, and MCP servers
- Keywords, author, repository, and license metadata
New skills (SKILL.md with YAML frontmatter):
- deep-research: systematic investigation with citations
- brainstorm: Socratic discovery and creative problem-solving
- troubleshoot: root cause analysis protocol (never retry blindly)
- pm: PDCA cycles, confidence checks, context persistence
- token-efficiency: ultra-compressed output mode
Enhanced hooks (hooks.json):
- SessionStart: initialize session context via script
- Stop: check for uncommitted changes and incomplete tasks
- PostToolUse (Write|Edit): verify edit correctness
MCP servers (.mcp.json):
- Context7: official library documentation
- Sequential Thinking: multi-step problem solving
Cleanup:
- Remove Python __init__.py files (not needed for Claude Code plugins)
- Add plugin README with installation and usage docs
Usage:
claude --plugin-dir ./plugins/superclaude # Local dev
/superclaude:research <topic> # Use a command
/superclaude:brainstorm <idea> # Use a skill
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
---------
Signed-off-by: Mithun Gowda B <mithungowda.b7411@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
* fix: fill implementation gaps across core modules
- Replace ConfidenceChecker placeholder methods with real implementations
that search the codebase for duplicates, verify architecture docs exist,
check research references, and validate root cause specificity
- Fix intelligent_execute() error capture: collect actual errors from
failed tasks instead of hardcoded None, format tracebacks as strings,
and fix variable shadowing bug where loop var overwrote task parameter
- Implement ReflexionPattern mindbase integration via HTTP API with
graceful fallback when service is unavailable
- Fix .gitignore: remove duplicate entries, add explicit !-rules for
.claude/settings.json and .claude/skills/, remove Tests/ ignore
- Remove unnecessary sys.path hack in cli/main.py
- Fix FailureEntry.from_dict to not mutate input dict
- Add comprehensive execution module tests: 62 new tests covering
ParallelExecutor, ReflectionEngine, SelfCorrectionEngine, and the
intelligent_execute orchestrator (136 total, all passing)
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: include test-generated reflexion artifacts
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* fix: address 5 open GitHub issues (#536, #537, #531, #517, #534)
Security fixes:
- #536: Remove shell=True and user-controlled $SHELL from _run_command()
to prevent arbitrary code execution. Use direct list-based subprocess.run
without passing full os.environ to child processes.
- #537: Add SHA-256 integrity verification for downloaded docker-compose
and mcp-config files. Downloads are deleted on hash mismatch. Gateway
config supports pinned hashes via docker_compose_sha256/mcp_config_sha256.
Bug fixes:
- #531: Add agent file installation to `superclaude install` and `update`
commands. 20 agent markdown files are now copied to ~/.claude/agents/
alongside command installation.
- #517: Fix MCP env var flag from --env to -e for API key passthrough,
matching the Claude CLI's expected format.
Usability:
- #534: Replace Japanese trigger phrases and report labels in pm-agent.md
and pm.md (both src/ and plugins/) with English equivalents for
international accessibility.
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* docs: align documentation with Claude Code and fix version/count gaps
- Update CLAUDE.md project structure to include agents/ (20 agents),
modes/ (7 modes), commands/ (30 commands), skills/, hooks/, mcp/,
and core/ directories. Add Claude Code integration points section.
- Fix version references: 4.1.5 -> 4.2.0 in installation.md,
quick-start.md, and package.json (was 4.1.7)
- Fix feature counts across all docs:
- Commands: 21 -> 30
- Agents: 14/16 -> 20
- Modes: 6 -> 7
- MCP Servers: 6 -> 8
- Update README.md agent count from 16 to 20
- Add docs/user-guide/claude-code-integration.md explaining how
SuperClaude maps to Claude Code's native features (commands,
agents, hooks, skills, settings, MCP servers, pytest plugin)
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: update test-generated reflexion log
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* docs: comprehensive Claude Code gap analysis and integration guide
- Rewrite docs/user-guide/claude-code-integration.md with full feature
mapping: all 28 hook events, skills system with YAML frontmatter,
5 settings scopes, permission rules, plan mode, extended thinking,
agent teams, voice, desktop features, and session management.
Includes detailed gap table showing where SuperClaude under-uses
Claude Code capabilities (skills migration, hooks integration,
plan mode, settings profiles).
- Add Claude Code native features section to CLAUDE.md with extension
points we use vs should use more (hooks, skills, plan mode, settings)
- Add Claude Code integration gap analysis to KNOWLEDGE.md with
prioritized action items for skills migration, hooks leverage,
plan mode integration, and settings profiles
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: update test-generated reflexion log
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* chore: bump version to 4.3.0
Bump version across all 15 files:
- VERSION, pyproject.toml, package.json
- src/superclaude/__init__.py, src/superclaude/__version__.py
- CLAUDE.md, PLANNING.md, TASK.md, CHANGELOG.md
- README.md, README-zh.md, README-ja.md, README-kr.md
- docs/getting-started/installation.md, quick-start.md
- docs/Development/pm-agent-integration.md
Also fixes __version__.py which was out of sync at 0.4.0.
Adds comprehensive CHANGELOG entry for v4.3.0.
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
* i18n: replace all Japanese/Chinese text with English in source files
Replace CJK text with English across all non-translation files:
- src/superclaude/commands/pm.md: 38 Japanese strings in PDCA cycle,
error handling patterns, anti-patterns, document templates
- src/superclaude/agents/pm-agent.md: 20 Japanese strings in PDCA
phases, self-evaluation, documentation sections
- plugins/superclaude/: synced from src/ copies
- .github/workflows/readme-quality-check.yml: all Chinese comments,
table headers, report strings, and PR comment text
- .github/workflows/pull-sync-framework.yml: Japanese comment
- .github/PULL_REQUEST_TEMPLATE.md: complete rewrite from Japanese
Translation files (README-ja.md, docs/user-guide-jp/, etc.) are
intentionally kept in their respective languages.
https://claude.ai/code/session_01AnGJMAA6Qp2j9WKKHHZfB9
---------
Co-authored-by: Claude <noreply@anthropic.com>
The pull-sync-framework workflow previously pushed directly to main on a
6-hour cron schedule with no human review gate. If the upstream framework
repo were compromised, malicious code would propagate automatically to
all downstream forks.
This change:
- Creates a dated branch (framework-sync/<date>) instead of pushing to main
- Opens a pull request for human review before merge
- Adds pull-requests: write permission for gh pr create
Fixes#535
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Fix AIRIS MCP Gateway installer: missing config, wrong SSE registration
Three bugs caused the gateway to be non-functional after installation:
1. mcp-config.json never created: The installer downloads docker-compose.yml
but never creates the mcp-config.json that it bind-mounts into the
container. Docker creates an empty file, so the gateway starts with
zero backend servers configured.
2. Wrong claude mcp add format for SSE transport: Used
`claude mcp add --transport sse name -- npx mcp-remote URL` which is
stdio syntax. Claude Code parsed "npx" as the SSE URL, producing
`{"type":"sse","url":"npx"}`. Fixed to pass the URL directly:
`claude mcp add --transport sse name URL`.
3. Template has airis-agent enabled as HOT but no container exists:
The downloaded mcp-config.template.json enables airis-agent (mode:hot)
but docker-compose.dist.yml has no airis-agent service. The installer
now disables airis-agent after downloading the template.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Fix AIRIS gateway installer: .env creation, mindbase disable, health check
- Create .env file with expanded HOST_WORKSPACE_DIR (no tilde) to fix
Docker Compose volume mount failures
- Disable mindbase in downloaded mcp-config.json (wrong subdirectory
path crashes gateway at startup)
- Add post-startup health check polling (6 attempts, 5s intervals)
- Update dry_run output to mention .env file creation
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style: fix ruff formatting for install_mcp.py
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Yuxin Liu <yuxinliu@YuxindeMac-mini.local>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
- Add missing [4.1.9] section with Tavily, Chrome DevTools, framework restoration
- Update [4.2.0] with actual changes (AIRIS Gateway, boundaries, command docs)
- Remove outdated Deep Research System placeholder (was from rolled-back PR #380)
- Update version badges and references in all README files
- Sync CLAUDE.md and PLANNING.md version references
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Add CRITICAL BOUNDARIES sections to 14 command files to clearly separate
document-only commands from execution commands. This prevents commands
from overstepping their intended scope.
Document-Only Commands (now explicitly STOP after output):
- brainstorm: requirements discovery only, no architecture
- workflow: implementation plan only, no code execution
- spawn: task decomposition only, delegates to other commands
- research: research report only, no implementation
- estimate: estimation report only, no execution
- troubleshoot: diagnosis by default, --fix flag required for fixes
- business-panel: expert analysis only, no implementation
- task: discrete execution, stops when complete
Execution Commands (now have clear completion criteria):
- implement: added completion checklist + handoff to /sc:test
- improve: added auto-fix vs approval-required thresholds
- cleanup: added safety thresholds for code removal
Handoff Additions (already correct, added Next Step):
- design: handoff to /sc:implement
- analyze: handoff to /sc:improve or /sc:cleanup
- spec-panel: handoff to /sc:design or /sc:implement
Documentation:
- commands.md: added "Command Output Categories" section
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
docs: add Commands Reference to docs/user-guide/commands.md
- Complete reference for all 30 /sc:* commands with syntax, examples,
workflows, and decision guides
- Moved from SC_COMMANDS_REFERENCE.md (root) to docs/user-guide/commands.md
- Updated README.md links to point to new location
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* docs: add AIRIS MCP Gateway as unified installation option
Adds airis-mcp-gateway as an alternative MCP setup method that provides:
- Single SSE endpoint for all MCP servers
- HOT/COLD server management for token optimization
- Lazy loading (servers start on-demand)
- Unified tool discovery
- Web UI for management
This addresses the complexity of managing multiple MCP servers and
provides token-efficient operation through selective tool advertising.
Repository: github.com/agiletec-inc/airis-mcp-gateway
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(docs): update AIRIS MCP Gateway installation instructions
- Replace outdated docker-compose.dist.yml method with quick-install.sh
- Fix Claude Code registration command: use --transport sse instead of npx mcp-remote
- Update feature list: 27+ tools, 75-90% token reduction, 120s idle timeout
- Add verification commands (health check, tools count)
- Update HOT/COLD server examples with actual server names
- Add AIRIS Suite reference (airis-agent, mindbase, airis-workspace)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(mcp): update airis configs to recommend airis-mcp-gateway
- Update mcp-servers.md: remove quick-install.sh, use git clone + docker compose
- Update MCP_Airis-Agent.md: recommend gateway, list current tools
- Update MCP_Mindbase.md: recommend gateway, list current tools
- Add deprecation notice to airis-agent.json and mindbase.json configs
- Change airis-agent config from Docker to uvx
All airis components (airis-agent, mindbase) are now included in airis-mcp-gateway.
Individual installation is deprecated in favor of the unified gateway.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat(mcp): add AIRIS MCP Gateway as default installation option
- Add AIRIS_GATEWAY config with endpoint and docker-compose URL
- Add check_docker_available() for Docker prerequisite check
- Add install_airis_gateway() for automated gateway installation
- Downloads docker-compose.dist.yml
- Starts containers with docker compose
- Registers SSE endpoint with Claude Code
- Update interactive selection to recommend gateway first (default: 'g')
- Update list_available_servers() to show gateway prominently
- Keep individual servers as legacy fallback option
Usage:
superclaude mcp # Interactive (gateway default)
superclaude mcp --servers airis-mcp-gateway # Direct install
superclaude mcp --servers tavily # Individual server
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* docs(claude): update CLAUDE.md with MCP gateway and cleanup
- Fix version: 0.4.0 → 4.1.9
- Add MCP server commands to Essential Commands section
- Update MCP section to recommend airis-mcp-gateway as default
- Consolidate v5.0 plugin warnings (3 locations → 1)
- Remove unused plugins/superclaude/ structure from docs
- Reduce file size: 312 → 284 lines
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* docs: add capability-mapping-v5 design document
7-verb capability model to reduce cognitive load:
- search, summarize, retrieve, plan, edit, execute, record
- Collapses 30 commands into 7 canonical capabilities
- Intent-based routing via airis-mcp-gateway
- Migration path: v4 commands → v5 capabilities
Addresses #501 (too many slash commands)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* style: fix ruff formatting in install_mcp.py
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(mcp): install gateway to ~/.superclaude/ instead of cwd
Avoids polluting the host by:
- Installing to ~/.superclaude/airis-mcp-gateway/
- Using --project-directory for docker compose
- Showing installation path in output
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: kazuki <kazuki@kazukinoMacBook-Air.local>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* docs: add AIRIS MCP Gateway as unified installation option
Adds airis-mcp-gateway as an alternative MCP setup method that provides:
- Single SSE endpoint for all MCP servers
- HOT/COLD server management for token optimization
- Lazy loading (servers start on-demand)
- Unified tool discovery
- Web UI for management
This addresses the complexity of managing multiple MCP servers and
provides token-efficient operation through selective tool advertising.
Repository: github.com/agiletec-inc/airis-mcp-gateway
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(docs): update AIRIS MCP Gateway installation instructions
- Replace outdated docker-compose.dist.yml method with quick-install.sh
- Fix Claude Code registration command: use --transport sse instead of npx mcp-remote
- Update feature list: 27+ tools, 75-90% token reduction, 120s idle timeout
- Add verification commands (health check, tools count)
- Update HOT/COLD server examples with actual server names
- Add AIRIS Suite reference (airis-agent, mindbase, airis-workspace)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix(mcp): update airis configs to recommend airis-mcp-gateway
- Update mcp-servers.md: remove quick-install.sh, use git clone + docker compose
- Update MCP_Airis-Agent.md: recommend gateway, list current tools
- Update MCP_Mindbase.md: recommend gateway, list current tools
- Add deprecation notice to airis-agent.json and mindbase.json configs
- Change airis-agent config from Docker to uvx
All airis components (airis-agent, mindbase) are now included in airis-mcp-gateway.
Individual installation is deprecated in favor of the unified gateway.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: kazuki <kazuki@kazukinoMacBook-Air.local>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* feat(mcp): add Airis Agent and MindBase MCP servers
Add Docker-based MCP server integrations:
- airis-agent: Confidence checking, deep research, repo indexing
- Prevents wrong-direction work with pre-implementation validation
- ghcr.io/agiletec-inc/airis-agent:latest
- mindbase: Semantic memory with pgvector embeddings
- Store/search memories with automatic embedding generation
- ghcr.io/kazuph/mindbase-mcp:latest
Both servers run via Docker for easy setup - no local dependencies.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* fix: use correct ghcr.io org for mindbase-mcp
---------
Co-authored-by: kazuki <kazuki@kazukinoMacBook-Air.local>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
The repository had both uppercase (e.g., docs/User-Guide/) and lowercase
(e.g., docs/user-guide/) paths tracked, causing persistent dirty state
on case-insensitive filesystems (macOS, Windows).
This commit removes the uppercase variants and keeps only lowercase paths,
which is the conventional standard for documentation directories.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add new test error solutions to solutions_learned.jsonl
- Document test failures for reflexion pattern testing
- Track database connection and unknown test errors
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Use force-include to add src/ as superclaude/_src/
- Use force-include to add plugins/ as superclaude/_plugins/
- Total package now has 343 files:
- Core superclaude/: 85 files
- _src/: 148 files (full source structure)
- _plugins/: 110 files (plugin resources)
- No share/ folder needed - everything in package
- Maintains backward compatibility
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add plugins as shared-data in wheel configuration
- Plugins installed to {prefix}/share/superclaude/plugins/
- Increased wheel files from 128 to 225 (110 plugin files)
- Maintains backward compatibility with existing package structure
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove redundant include patterns that were causing mis-matched data size
- Keep only plugins/** patterns in wheel.include
- Packages in src/ are automatically included by hatchling
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Bump version from 4.1.8 to 4.1.9 in all relevant files
- Add __init__.py to all directories in src/superclaude/ and plugins/superclaude/
- Update pyproject.toml to include plugins/superclaude package
- Update MANIFEST.in to include plugin resources
- Ensure proper Python package structure with 27 __init__.py files
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* feat: add missing install.sh script referenced in README\n\n- Create comprehensive installation script with POSIX compatibility\n- Add interactive and non-interactive installation modes\n- Include prerequisites checking and MCP server setup guidance\n- Replace echo -e with printf for better POSIX compliance
* fix: resolve linting errors in install_mcp.py and clean_command_names.py
Fix multiple ruff linting errors to ensure CI/CD pipeline passes:
- install_mcp.py: Remove unused pathlib.Path import, replace bare except
with specific exception types (ValueError, IndexError), remove
extraneous f-string prefixes on lines without placeholders
- clean_command_names.py: Remove unused os import, convert f-strings
without placeholders to regular strings
- pyproject.toml: Exclude docs/ directory from ruff checks to avoid
N999 module naming violations in documentation templates
All linting checks now pass successfully.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* style: apply ruff format to Python source files
Apply ruff formatting rules to CLI and scripts modules to ensure
consistent code style across the codebase.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(ci): remove incompatible pip cache from quick-check workflow
## Problem
GitHub Actions was failing with error:
"Cache folder path is retrieved for pip but doesn't exist on disk:
/home/runner/.cache/pip. This likely indicates that there are no
dependencies to cache."
## Root Cause
The quick-check.yml workflow specified `cache: 'pip'` in the Python
setup step, but the workflow uses UV (not pip) for package management
via `uv pip install --system -e ".[dev]"`.
UV uses its own cache directory (~/.cache/uv), so the pip cache path
was never created, causing the error.
This was a migration oversight:
- When UV was adopted as the project standard (commit 00706f0), the
CLAUDE.md established "CRITICAL: Never use pip directly" rule
- The test.yml workflow was created correctly without pip cache
- The quick-check.yml workflow incorrectly included pip cache from
initial creation (commit 8c0559c) and was not updated during migration
## Solution
Remove `cache: 'pip'` line to align with:
- Project's UV-first architecture (CLAUDE.md)
- test.yml workflow (which runs successfully without pip cache)
- readme-quality-check.yml workflow (no cache needed)
Note: publish-pypi.yml intentionally uses pip cache as it directly
runs `python -m pip install` commands, which is correct for that workflow.
## Impact
- ✅ Eliminates GitHub Actions cache warning
- ✅ Aligns all UV-based workflows consistently
- ✅ Follows project standards documented in CLAUDE.md
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
Fixed path resolution in _get_commands_source() to check package
location first before falling back to plugins/ directory.
Changes:
- Priority 1: package_root/commands/ (installed package via pipx/pip)
- Priority 2: plugins/superclaude/commands/ (source checkout)
- Return package location on failure (clearer error message)
This fixes the error:
"Command source directory not found: /path/to/.local/pipx/venvs/
superclaude/lib/python3.13/plugins/superclaude/commands"
The function was trying plugins/ path first and returning it even
when it didn't exist. Now it correctly prioritizes the installed
package location where commands are actually bundled.
Note: install_skill.py already had correct logic and doesn't need changes.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Added explicit entries for commands and skills directories to make it
clear these are included in the source distribution.
While these were already covered by the recursive-include for
src/superclaude/*.md, explicit entries improve maintainability
and documentation.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fixed "Command source directory not found" error when installing via pipx.
Changes:
- Copied command files from plugins/superclaude/commands/ to src/superclaude/commands/
- Updated pyproject.toml to include commands in wheel package
- Added README.md explaining command sync requirement
- install_commands.py already had fallback logic to check package_root/commands/
The commands directory will be included in the installed package, allowing
'superclaude install' to work correctly from pipx installations.
Note: In v4.x, commands must be synced between plugins/ and src/superclaude/.
In v5.0, the plugin system will use plugins/ directly.
Fixes installation error:
Command source directory not found: /path/to/plugins/superclaude/commands
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
kazuki