Scripts, screenshots, and reports are part of a project's evolution — save
them to .claude/scripts/, .claude/screenshots/, .claude/artifacts/ instead
of /tmp where they're lost. Scripts are committed; screenshots/artifacts
are gitignored. Updated gitignore templates, agent prompts, and skill-creator.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Skills that scan files or collect data should instruct Claude to write a
script rather than chain many bash commands — reduces approval prompts
from N to 2 (one Write + one Bash). Also updated project-health agent
prompts to prefer Read/Glob/Grep and scripted analysis.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add commands/project-health.md as slash command (/project-health)
- Remove skills/project-kickoff/ (merged into project-health)
- Remove skills/context-manager/ (merged into project-health)
- Update plugin.json description
- project-health skill retains all reference files from both old skills
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Added: cd, rm, rmdir, ln, readlink, realpath, stat, file, chown,
gunzip, yq, cut, tr, uniq, xargs, basename, dirname, kill, export,
source, id, whoami, hostname, uptime.
These are all commands that could trigger permission prompts during
normal development. The goal is zero prompts after project setup.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New unified skill that handles the full project config lifecycle:
- Setup: permissions, CLAUDE.md, .gitignore from scratch
- Health check: audit permissions + context + docs in one pass
- Tidy: clean leaked secrets, cruft, legacy syntax
- Capture: save session learnings to CLAUDE.md
- Add preset: merge new permission presets
- Restructure: split oversized files, resolve overlaps
Key changes from predecessors:
- Sub-agents for heavy analysis (keeps main context clean)
- No Python scripts (replaced by Task agents using native tools)
- Merged templates from both skills into one reference file
- MCP auto-discovery via ToolSearch (not hardcoded server names)
- Documents all permission wildcard findings (mcp__* doesn't work)
Old skills (project-kickoff, context-manager) left in place for now.
Remove once project-health is validated.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Goal: never see a permission prompt after project setup
- Auto-discover MCP servers via ToolSearch (can't blanket wildcard)
- Document what does NOT work: mcp__*, mcp__*__*, gh * subcommand bug
- Always include Universal Base + detected language + deployment presets
- Include more rather than less — unused permissions cost nothing
- Document hot-reload limitation and shadow-not-merge behaviour
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
gh issue, gh pr, etc. sometimes don't match Bash(gh *) alone.
Adding explicit subcommand patterns as workaround.
Also confirmed mcp__*__* does NOT work for blanket MCP permissions.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Discovered that mcp__* does NOT match across the __ boundary.
You must use mcp__servername__* for each server individually.
Updated docs and removed incorrect blanket recommendation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replaces PR #40 with a focused version in plugins/dev-tools/. Tests layout
responsiveness only (not general accessibility), uses single-session resize
instead of per-breakpoint sub-agents, and detects layout transitions between
adjacent viewport widths.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
project-kickoff: Bootstrap new projects with curated permission presets
(cloudflare, vercel, node, python, ops/admin), CLAUDE.md templates, and
.gitignore. Includes tidy_permissions.py to analyse messy settings files
— detects leaked secrets, shell fragments, deprecated MCP refs, and
consolidation opportunities.
gemini-guide: Documentation lookup skill for Gemini API — routes queries
to local cached docs (~/Documents/google-gemini-context/) and provides
correction tables for deprecated SDK patterns. Trial of the "guide agent"
pattern (brings docs to Claude, vs gemini-peer-review which sends code
to Gemini).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move all 24 skills from flat skills/ directory into 7 category-based
plugins under plugins/. Add new Shopify plugin with 3 skills
(shopify-setup, shopify-products, shopify-content) for Admin API
access, product CRUD/bulk import, and content management.
Update marketplace.json to new plugin format, fix script paths in
skill-creator, context-manager, and vite-flare-starter SKILL.md files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Jeremy Dawes