Drop scheduled routines and every skill the setup does not use

Routines that watch one site's CI and deploys are that project's automation,
not setup config. Shipping them here meant a portable bundle carried one
person's monitoring jobs, and installing the bundle overwrote whatever routines
the machine already had. They belong next to the project they watch. The
launchd templates and the install and verify steps that served them go too.

Seven skills are deleted rather than shipped disabled. A skill that is not
worth loading is not worth carrying, and "installed but off" is a state that
invites re-enabling something already covered elsewhere:
verification-before-completion, systematic-debugging, requesting-code-review,
tokenmaxxing, vercel-deploy, spaceship, security-threat-model. ATTRIBUTION.md
records why each one went. Their skillOverrides entries go with them, since an
override naming a skill that no longer exists is dead config.

22 skills remain, all of them in use.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: itsvedantkumar

.claude/verify.sh

@@ -32,18 +32,6 @@ else
   skip "json valid" "jq not installed"
 fi
 
-# --- 3. launchd templates are valid plists --------------------------------------------------
-if command -v plutil >/dev/null; then
-  errs=""
-  for t in launchd/*.plist.tmpl; do
-    [ -e "$t" ] || continue
-    plutil -lint "$t" >/dev/null 2>&1 || errs="$errs\n$t"
-  done
-  [ -z "$errs" ] && ok "launchd templates" || bad "launchd templates" "$(printf '%b' "$errs")"
-else
-  skip "launchd templates" "plutil not available (not macOS)"
-fi
-
 # --- 4. skills are loadable ------------------------------------------------------------------
 # A skill with no description, or one longer than the configured listing cap, gets truncated
 # out of the listing and silently stops auto-triggering. That is the failure this catches.
@@ -91,10 +79,6 @@ hits=$(grep -rIn --exclude-dir=.git -E \
   . 2>/dev/null | head -5)
 [ -z "$hits" ] && ok "no committed secrets" || bad "no committed secrets" "$hits"
 
-# --- 7. routines use gh, never scraped tokens or absent tools ----------------------------------
-hits=$(grep -rIln -e 'github_pat_' -e 'Vercel MCP' -e 'remote get-url.*sed' claude/scheduled-tasks 2>/dev/null)
-[ -z "$hits" ] && ok "routines use gh auth" || bad "routines use gh auth" "$hits"
-
 # --- 8. no infrastructure identifiers ---------------------------------------------------------
 # This repo is public and its routines are templates. Real Vercel project or team IDs, Claude
 # Code environment IDs, and cloud routine trigger IDs identify live infrastructure, so they

README.md

@@ -1,11 +1,11 @@
 # vstack
 
 One repo that installs a complete Claude Code setup on a Mac: skills that fire on their own,
-subagents, hooks, MCP servers, deploy scripts, and scheduled routines.
+subagents, hooks, MCP servers, and deploy scripts.
 
-Everything here used to live in four places at once. Some of it sat in `~/.claude`, some in
-`~/.config/agents`, some only in a cloud routine, and the parts that mattered most were the
-parts nothing tracked. Reinstalling meant remembering. Now it is one clone and one command.
+Everything here used to live in three places at once: some in `~/.claude`, some in
+`~/.config/agents`, and the parts that mattered most in nothing at all. Reinstalling meant
+remembering. Now it is one clone and one command.
 
 ```bash
 git clone https://github.com/itsvedantkumar/vstack.git
@@ -55,13 +55,11 @@ Check a machine without changing it:
 
 | Component | Count | Installs to |
 |---|---|---|
-| Skills | 29 (24 active, 5 off) | `~/.claude/skills/` |
+| Skills | 22 | `~/.claude/skills/` |
 | Subagents | 7 | `~/.claude/agents/` |
 | Slash commands | 15 | `~/.claude/commands/` |
 | Hooks | 4 | `~/.claude/hooks/` |
 | CLI wrappers | 8 | `~/.config/agents/bin/` |
-| Scheduled routines | 3 | `~/.claude/scheduled-tasks/` |
-| launchd timers | 3 | `~/Library/LaunchAgents/` (only with `--with-launchd`) |
 | MCP servers | 2 | merged into `~/.claude.json` |
 
 Regenerate these counts with `find claude/skills -maxdepth 1 -mindepth 1 -type d | wc -l` and
@@ -92,8 +90,8 @@ back as the reason. Three blocks per session, then it stops, so an overnight run
 forever.
 
 This repo gates itself with the same mechanism. Run `./.claude/verify.sh` and it checks shell
-syntax, JSON validity, plist validity, skill frontmatter and description lengths, hardcoded
-home paths, committed credentials, and a full `install.sh --dry-run`. To add the same gate to
+syntax, JSON validity, skill frontmatter and description lengths, hardcoded home paths,
+committed credentials, infrastructure identifiers, and a full `install.sh --dry-run`. To add the same gate to
 another repo, ask Claude for verification in that repo and the `create-verification-skill`
 skill writes one that fits its stack.
 
@@ -107,20 +105,6 @@ The example deliberately names the Anthropic key `ANTHROPIC_SDK_API_KEY`. Claude
 `ANTHROPIC_API_KEY` and bills API credits against it instead of your subscription, so the
 shell wrapper strips that name from the environment and `doctor` fails if it is set.
 
-## Scheduling
-
-The three routines under `claude/scheduled-tasks/` are prompts. Scheduling them is a separate
-choice, and there are two lanes:
-
-- **Cloud routines** at [claude.ai/code/routines](https://claude.ai/code/routines) run whether
-  or not the Mac is awake. This is the better default.
-- **launchd timers** run locally. Install them with `./install.sh --with-launchd`.
-
-Pick one lane. Running both doubles every job.
-
-The three routines are templates, not live jobs. They carry `<owner>/<repo>` and
-`prj_YOUR_PROJECT_ID` placeholders, and each file opens with the list of values to replace.
-`./.claude/verify.sh` fails if a real Vercel, environment, or trigger ID lands in the repo.
 
 ## Health check
 
@@ -135,11 +119,10 @@ Claude Code update: plugin updates have reverted config here before.
 ## Layout
 
 ```
-claude/          settings, hooks, agents, commands, skills, scheduled-tasks
+claude/          settings, hooks, agents, commands, skills
 bin/             CLI wrappers installed to ~/.config/agents/bin
 shell/           zsh wrapper and env snippet
 mcp/             MCP server definitions merged into ~/.claude.json
-launchd/         plist templates, __HOME__ substituted at install time
 install.sh       user-scope install, idempotent
 overlay.sh       copies the config into a repo so cloud sessions get it
 ```
@@ -171,11 +154,9 @@ directory, so a committed `.claude/` directory is the only config it can read. R
 
 ## Credits
 
-The 29 skills come from four places: 18 ported from [pstack](https://github.com/cursor/plugins),
-7 from [Superpowers](https://github.com/obra/superpowers), 2 third-party (Apache 2.0 and MIT),
-and 2 written for this setup. Per-skill sources, licenses, and the reason each disabled skill
-is disabled are in `claude/skills/ATTRIBUTION.md`.
+The 22 skills come from two places: 18 ported from [pstack](https://github.com/cursor/plugins)
+and 4 from [Superpowers](https://github.com/obra/superpowers). Sources and licenses are
+recorded in `claude/skills/ATTRIBUTION.md`.
 
-Five skills ship disabled through `skillOverrides` in `claude/settings.json`. They are kept
-rather than deleted because the originals live nowhere else, and a disabled skill costs no
-context. That file is the one place that decides what is active.
+Every skill here is one this setup actually uses. A skill that turned out to be redundant was
+deleted rather than shipped disabled: if it is not worth loading, it is not worth carrying.

bin/spaceship-mcp

@@ -8,12 +8,9 @@
     "xlsx": "off",
     "docx": "off",
     "pptx": "off",
-    "tokenmaxxing": "off",
     "build-mcp-server": "off",
-    "vercel-deploy": "off",
     "dataviz": "name-only",
     "pdf": "name-only",
-    "spaceship": "name-only",
     "skill-creator": "name-only",
     "update-config": "name-only",
     "keybindings-help": "name-only",
@@ -25,9 +22,6 @@
     "run": "name-only",
     "simplify": "name-only",
     "security-review": "name-only",
-    "security-threat-model": "name-only",
-    "systematic-debugging": "off",
-    "requesting-code-review": "off",
     "frontend-design:frontend-design": "name-only",
     "claude-mem:babysit": "off",
     "claude-mem:cloud-sync": "off",
@@ -47,8 +41,7 @@
     "claude-mem:what-the": "off",
     "claude-mem:wowerpoint": "off",
     "claude-mem:mem-search": "name-only",
-    "claude-mem:smart-explore": "name-only",
-    "verification-before-completion": "off"
+    "claude-mem:smart-explore": "name-only"
   },
   "precomputeCompactionEnabled": true,
   "promptSuggestionEnabled": false,