get-claudia 1.60.1 → 1.61.0

package/CHANGELOG.md

@@ -2,6 +2,24 @@
 
 All notable changes to Claudia will be documented in this file.
 
+## 1.61.0 (2026-05-22)
+
+### Launch from anywhere, upgrade from anywhere
+
+This release adds a `claudia` shell command (and `update-claudia`) so you no longer have to remember where you installed Claudia or `cd` there before launching or upgrading. New installs and upgrades wire the helper into `~/.zshrc` and `~/.bashrc` automatically. After upgrading once via `npx get-claudia .`, future upgrades are just `update-claudia` from any directory.
+
+### Added
+
+- **`claudia` shell command for launching from anywhere.** New installs (and upgrades) get a `claudia` shell function written to `~/.claudia/shell-init.sh` and sourced from `~/.zshrc` and `~/.bashrc`. From any directory, `claudia` cd's to your install folder and launches `claude`. All `claude` flags pass through (`claudia --resume`, `claudia -c "morning brief"`, etc). The folder path is stored in `~/.claudia/claudia-home`; edit it if you move your install. The installer is idempotent: re-running won't double-add to your rc files. The shell function is a smart router that delegates known npm-CLI subcommands (`setup`, `system-health`, `google`, `doctor`) back to the `claudia` binary when it's on PATH, so existing workflows keep working.
+- **`claudia yolo` subcommand.** Sugar for `claude --dangerously-skip-permissions` (cd's first, then passes through any additional flags). Same routing logic — `claudia yolo --resume` works.
+- **`update-claudia` command (also `claudia update`).** Runs `npx get-claudia <your-install-path>` from any directory by reading the stored path from `~/.claudia/claudia-home`. No more "wait, where did I install Claudia?" before upgrading. Both forms are aliases for the same logic.
+- **`~/.claudia/claudia-home`.** Single-line file with the absolute path of the user's Claudia install. Written by the installer; read by the shell function. Move your install? Edit this file.
+
+### Fixed
+
+- **ASCII greeting logo: row 1 (hair top) realigned over face** (in `template-v2/CLAUDE.md`). The top hair row was indented 6 leading spaces while the face row below it starts at column 0, so the hair appeared shifted ~4 characters to the right. Now sits correctly above the face. Existing installs see the fix on next upgrade.
+- **Installer no longer treats flag-looking arguments as install targets.** Previously, `npx get-claudia --help` (or any unrecognized `--flag`) was interpreted as the install path and would create a `./--help/` directory with templates copied into it. Now: `--help`/`-h` prints a proper usage message and exits, `--version`/`-V` prints the version, and any other leading-dash argument prints `Error: unrecognized argument: <flag>` and exits 1 without touching the filesystem.
+
 ## 1.60.1 (2026-05-22)
 
 ### Two bug fixes from the fixing-phase pass

package/bin/index.js

@@ -14,6 +14,7 @@ import {
   resolveBakPath,
   applyResolution,
 } from './manifest-lib.js';
+import { writeShellInit, appendShellRC } from './shell-init.js';
 
 const __filename = fileURLToPath(import.meta.url);
 const __dirname = dirname(__filename);
@@ -63,6 +64,30 @@ function getVersion() {
   }
 }
 
+function printUsage(version) {
+  console.log(`get-claudia v${version}
+
+Usage:
+  npx get-claudia [target-dir]   Install or upgrade Claudia (default: ./claudia)
+  npx get-claudia .              Install or upgrade in the current directory
+  npx get-claudia upgrade        Same as \`.\` (in-place upgrade)
+  npx get-claudia google         Set up Google Workspace integration
+
+Flags:
+  --skip-memory                  Skip memory daemon setup
+  --dev                          Dev mode (load daemon from local source)
+  --yes, -y                      Non-interactive (auto-confirm prompts)
+  --help, -h                     Show this help and exit
+  --version, -V                  Print version and exit
+
+After install, from any directory:
+  claudia                        cd to your install folder, launch claude
+  claudia yolo                   Launch claude --dangerously-skip-permissions
+  claudia update                 Upgrade Claudia (no need to cd first)
+  update-claudia                 Alias for \`claudia update\`
+`);
+}
+
 // Simple y/n prompt. Returns true if user confirms (or non-TTY / --yes flag).
 function confirm(question) {
   if (!isTTY || process.argv.includes('--yes') || process.argv.includes('-y')) {
@@ -284,6 +309,7 @@ const STEPS = [
   { id: 'memory',      label: 'Memory System' },
   { id: 'daemon',      label: 'Memory Daemon' },
   { id: 'mcp',         label: 'MCP Config' },
+  { id: 'shell',       label: 'Shell Helper' },
   { id: 'vault',       label: 'Obsidian Vault' },
   { id: 'health',      label: 'Health Check' },
 ];
@@ -737,6 +763,16 @@ async function checkForNewerVersion(currentVersion) {
 async function main() {
   const version = getVersion();
 
+  // Handle --help / --version BEFORE banner, update-check, or any I/O.
+  if (process.argv.includes('--help') || process.argv.includes('-h')) {
+    printUsage(version);
+    process.exit(0);
+  }
+  if (process.argv.includes('--version') || process.argv.includes('-V')) {
+    console.log(version);
+    process.exit(0);
+  }
+
   // Self-update trampoline: re-exec with latest if we're stale
   const newerVersion = await checkForNewerVersion(version);
   if (newerVersion) {
@@ -773,6 +809,16 @@ async function main() {
   const filteredArgs = args.filter(a => a !== '--no-memory' && a !== '--skip-memory' && a !== '--dev' && a !== '--yes' && a !== '-y');
   const arg = filteredArgs[0];
 
+  // Reject flag-looking arguments early so they don't get used as install paths
+  // (e.g. `npx get-claudia --foo` would otherwise create a ./--foo/ directory).
+  // --help and --version are handled at the top of main(); anything else with a
+  // leading dash is a typo or unsupported flag.
+  if (typeof arg === 'string' && arg.startsWith('-')) {
+    console.error(`${colors.red}Error:${colors.reset} unrecognized argument: ${arg}`);
+    console.error(`Run ${colors.cyan}npx get-claudia --help${colors.reset} for usage.`);
+    process.exit(1);
+  }
+
   // ─── Subcommand: get-claudia google ─────────────────────────────────────
   if (arg === 'google') {
     await runGoogleSetup();
@@ -1517,13 +1563,35 @@ async function main() {
 
   renderer.stopSpinner();
 
-  // Vault step, then completion
-  runVaultStep(renderer, () => {
-    renderer.render();
-    showDbScanResults(dbScan);
-    showCompletion(targetDir, isCurrentDir, memoryOk, rootCause, isUpgrade);
+  // Shell helper step, then vault, then completion
+  runShellStep(renderer, targetDir, () => {
+    runVaultStep(renderer, () => {
+      renderer.render();
+      showDbScanResults(dbScan);
+      showCompletion(targetDir, isCurrentDir, memoryOk, rootCause, isUpgrade);
+    });
   });
 
+  // ── Shell helper step ──
+
+  function runShellStep(renderer, targetPath, callback) {
+    renderer.update('shell', 'active', 'installing claudia command...');
+    try {
+      writeShellInit(homedir(), targetPath);
+      const rc = appendShellRC(homedir());
+      if (rc.skipped) {
+        renderer.update('shell', 'done', 'files written (Windows: source manually)');
+      } else if (rc.added.length > 0) {
+        renderer.update('shell', 'done', `added to ${rc.added.length} rc file(s)`);
+      } else {
+        renderer.update('shell', 'done', 'already installed');
+      }
+    } catch (err) {
+      renderer.update('shell', 'warn', `${err.message}`);
+    }
+    callback();
+  }
+
   // ── Vault step ──
 
   function runVaultStep(renderer, callback) {
@@ -1651,6 +1719,7 @@ async function main() {
         console.log('');
         console.log(`   ${colors.cyan}${launchCmd}${colors.reset}`);
         console.log('');
+        console.log(` ${colors.dim}Tip: open a new terminal and just type ${colors.reset}${colors.cyan}claudia${colors.reset}${colors.dim} from anywhere.${colors.reset}`);
         console.log(` ${colors.dim}What's new: /morning-brief · /inbox-check · /feedback${colors.reset}`);
       } else {
         // Fresh install: build anticipation for the onboarding
@@ -1661,6 +1730,7 @@ async function main() {
         }
         console.log(`   ${colors.cyan}claude${colors.reset}`);
         console.log('');
+        console.log(` ${colors.dim}Or open a new terminal and type ${colors.reset}${colors.cyan}claudia${colors.reset}${colors.dim} from anywhere.${colors.reset}`);
         console.log(` ${colors.dim}She'll introduce herself and learn how you work.${colors.reset}`);
         console.log(` ${colors.dim}Try: ${colors.reset}${colors.cyan}"Say hi"${colors.reset} ${colors.dim}·${colors.reset} ${colors.cyan}/morning-brief${colors.reset} ${colors.dim}·${colors.reset} ${colors.cyan}"Who do I know?"${colors.reset}`);
       }

package/bin/shell-init.js

@@ -0,0 +1,127 @@
+// Shell helper installer for `claudia` command.
+//
+// Writes two files into ~/.claudia/:
+//   - claudia-home      : single-line file with the absolute path to the user's
+//                         Claudia install directory (where `claude` should launch).
+//   - shell-init.sh     : defines the `claudia` shell function (and `claudia yolo`).
+//
+// Then idempotently appends a one-line source to ~/.zshrc and ~/.bashrc so the
+// function is available in every new shell. The marker comment is used to detect
+// existing installs and avoid double-adding.
+//
+// On Windows we only write the files; rc-file plumbing is a no-op since neither
+// zsh nor bash is standard there. The function content is still useful for users
+// running WSL or Git Bash who source it manually.
+
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'fs';
+import { join } from 'path';
+
+export const SHELL_INIT_MARKER = '# Claudia shell helpers';
+
+const RC_SNIPPET = `
+${SHELL_INIT_MARKER} (\`claudia\` from anywhere, \`claudia yolo\` skips permissions)
+[ -f "$HOME/.claudia/shell-init.sh" ] && source "$HOME/.claudia/shell-init.sh"
+`;
+
+export const SHELL_INIT_CONTENT = `# Claudia shell helpers — sourced from your shell rc.
+# Edit ~/.claudia/claudia-home to change which folder \`claudia\` launches from.
+
+_claudia_home() {
+  local home_file="$HOME/.claudia/claudia-home"
+  if [ ! -f "$home_file" ]; then
+    echo "Claudia home not configured. Run: npx get-claudia ." >&2
+    return 1
+  fi
+  local dir
+  dir="$(cat "$home_file")"
+  if [ ! -d "$dir" ]; then
+    echo "Claudia home directory not found: $dir" >&2
+    echo "Fix by editing $home_file" >&2
+    return 1
+  fi
+  printf '%s' "$dir"
+}
+
+_claudia_cd() {
+  local dir
+  dir="$(_claudia_home)" || return 1
+  cd "$dir"
+}
+
+update-claudia() {
+  local dir
+  dir="$(_claudia_home)" || return 1
+  echo "Updating Claudia at $dir ..."
+  npx get-claudia "$dir"
+}
+
+claudia() {
+  case "$1" in
+    yolo)
+      shift
+      _claudia_cd && claude --dangerously-skip-permissions "$@"
+      ;;
+    update)
+      shift
+      update-claudia "$@"
+      ;;
+    setup|system-health|google|doctor|--version|-V|help|--help|-h)
+      # Pass known npm-CLI subcommands through to the binary (if installed).
+      command claudia "$@"
+      ;;
+    *)
+      _claudia_cd && claude "$@"
+      ;;
+  esac
+}
+`;
+
+// Write ~/.claudia/claudia-home and ~/.claudia/shell-init.sh.
+// Returns { homeFile, initFile } absolute paths for caller logging.
+export function writeShellInit(homeDir, claudiaTargetDir) {
+  const claudiaConfigDir = join(homeDir, '.claudia');
+  mkdirSync(claudiaConfigDir, { recursive: true });
+
+  const homeFile = join(claudiaConfigDir, 'claudia-home');
+  const initFile = join(claudiaConfigDir, 'shell-init.sh');
+
+  writeFileSync(homeFile, `${claudiaTargetDir}\n`);
+  writeFileSync(initFile, SHELL_INIT_CONTENT);
+
+  return { homeFile, initFile };
+}
+
+// Idempotently append the source line to a single rc file. Creates the file if
+// it doesn't exist (the source line is harmless on its own). Returns one of:
+//   'added'     - the source line was just appended
+//   'unchanged' - marker already present, nothing written
+function appendToRc(rcPath) {
+  let existing = '';
+  if (existsSync(rcPath)) {
+    existing = readFileSync(rcPath, 'utf8');
+    if (existing.includes(SHELL_INIT_MARKER)) {
+      return 'unchanged';
+    }
+  }
+  // Ensure separation from prior content
+  const sep = existing.length === 0 || existing.endsWith('\n') ? '' : '\n';
+  writeFileSync(rcPath, existing + sep + RC_SNIPPET);
+  return 'added';
+}
+
+// Append to the user's zsh and bash rc files. Skips on Windows.
+// Returns { added: [...], unchanged: [...] } of rc paths.
+export function appendShellRC(homeDir, platform = process.platform) {
+  const result = { added: [], unchanged: [], skipped: false };
+  if (platform === 'win32') {
+    result.skipped = true;
+    return result;
+  }
+  const rcFiles = [join(homeDir, '.zshrc'), join(homeDir, '.bashrc')];
+  for (const rc of rcFiles) {
+    const status = appendToRc(rc);
+    if (status === 'added') result.added.push(rc);
+    else result.unchanged.push(rc);
+  }
+  return result;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "get-claudia",
-  "version": "1.60.1",
+  "version": "1.61.0",
   "description": "An AI assistant who learns how you work.",
   "keywords": [
     "claudia",

package/template-v2/.claude/manifest.json

@@ -1,6 +1,6 @@
 {
-  "version": "1.60.1",
-  "generated": "2026-05-23T01:39:21.155Z",
+  "version": "1.61.0",
+  "generated": "2026-05-23T02:25:12.290Z",
   "algorithm": "sha256",
   "files": {
     ".claude/rules/claudia-principles.md": "939e9720421628e7f2e4c8dfbaa4aeb9c1e18e8c6a5379cd6b772a6835b812e5",
@@ -26,7 +26,6 @@
     ".claude/skills/capture-meeting/SKILL.md": "815e7587fcd925323c379a6c57e458b3484d034338c1237378641d142c126175",
     ".claude/skills/capture-meeting/evals/basic.yaml": "47c4eb1479aeda8067bc877cebfe8688c9b2bb3ce09297e42375cdf558f412e3",
     ".claude/skills/client-health/SKILL.md": "fed639e6f2c4433cab6a6b4b59776985492dda16be3bd1e843c3e94175936655",
-    ".claude/skills/close-the-loop/SKILL.md": "835af2b0bde5583d1dffa07f261a1ec50e87f99eb39694c00cb7965ccce21942",
     ".claude/skills/commitment-detector.md": "66f9e919349d1880eb12636cec6f583be1b3fbe3b28f6180ec732d3c284af3b6",
     ".claude/skills/connector-discovery.md": "a6659f6ed8a182f968cdc9b4a342de9015d5579859310b77ff230ce73397cdc5",
     ".claude/skills/databases/SKILL.md": "b2ce4ae30aafa2487dabd7befd067976f0092fcef31dea7601e1683ceca0ef50",
@@ -46,7 +45,6 @@
     ".claude/skills/ingest-sources/SKILL.md": "938b93d299649db156ad19e9a7bd5d09e950f53bc2eaf8a85386f9fdbca350d0",
     ".claude/skills/ingest-sources/references/extraction-patterns.md": "28d6dc604c4a11eacfe4523e705ad430ba8fe26632bafcccaef656e9618e5aea",
     ".claude/skills/judgment-awareness.md": "0077129a87f91d295515108f3f9a68ca6874c6e38fc7f2a73a462704fa00ba2d",
-    ".claude/skills/manuscript-fact-check/SKILL.md": "3dbdeea105740e42b1ff9d5e7245dd37418401b36655169b5a5ab2a4a8d6a383",
     ".claude/skills/map-connections/SKILL.md": "e8133d5e3de36e32e3858a1ee3918a577855e77ade022933097f87a0100cc9a3",
     ".claude/skills/meditate/SKILL.md": "1d814f0c52f4309708669139ba4b637bf05a228e2381950ee3851c6145e25cc7",
     ".claude/skills/meditate/evals/basic.yaml": "daba441b2fd9d1d4afddcff6eaa9673884198b1d0f85d9d06edbe0738012291a",
@@ -78,6 +76,6 @@
     ".claude/skills/wiki/SKILL.md": "e45e8db6fce1d774e36c93be27e5d9153b3ae474c48db478a4d7ff2b138e42e8",
     ".claude/skills/wiki/references/citations-and-contradictions.md": "1b16301020fdb5e4840035ca22425ebb03202694e88eb494105ff39e22e03263",
     ".claude/skills/wiki/references/page-template.md": "b19382ee9366b1e7f8c9f230881ac3c45982b59ffee0d82fe64b103862ba4745",
-    "CLAUDE.md": "736c5d95f1477ea6ef1c00d3006eeb4b369eb2bd5f3a926e4d78c42f857fa881"
+    "CLAUDE.md": "b8244c281afd5cf46544f1f68b4130672d9f6e6bcbcbae7defca3a31c09d4c12"
   }
 }

package/template-v2/CLAUDE.md

@@ -104,7 +104,7 @@ Uses three shades to approximate the installer's coloring: `▓▓` = hair, `█
 
 ```
 
-      ▓▓▓▓▓▓▓▓▒▒
+  ▓▓▓▓▓▓▓▓▒▒
 ▓▓██████████▒▒
 ▓▓██  ██  ██▓▓
   ██████████
@@ -304,6 +304,27 @@ I use proactive, contextual, and explicit skills. See `.claude/skills/README.md`
 | My learnings about you | `context/learnings.md` |
 | Project details | `projects/[project]/overview.md` |
 | Filed documents | `~/.claudia/files/` (entity-routed) |
+| Shell helper init | `~/.claudia/shell-init.sh` (sourced from `.zshrc`/`.bashrc`) |
+| Install path pointer | `~/.claudia/claudia-home` (one line, absolute path) |
+
+---
+
+## Launching from the Terminal
+
+Once installed, the `claudia` shell command works from any directory:
+
+| You type | What happens |
+|----------|--------------|
+| `claudia` | cd to your Claudia folder, launch `claude` |
+| `claudia yolo` | cd, launch `claude --dangerously-skip-permissions` |
+| `claudia -c "morning brief"` | cd, launch `claude -c "morning brief"` (any `claude` flag passes through) |
+| `claudia --resume` | cd, launch `claude --resume` |
+| `claudia update` / `update-claudia` | runs `npx get-claudia <your-install-path>` from anywhere — no need to cd first |
+| `claudia setup` / `claudia system-health` | passes through to the npm `claudia` CLI binary |
+
+The folder location is stored in `~/.claudia/claudia-home`. Move your install? Edit that file. Re-run `npx get-claudia .` from a new location and it'll update automatically.
+
+On Windows the shell function is written to `~/.claudia/shell-init.sh` but not auto-sourced. WSL or Git Bash users can `source ~/.claudia/shell-init.sh` manually (or add it to their rc file).
 
 ---
 

package/template-v2/.claude/skills/close-the-loop/SKILL.md

@@ -1,139 +0,0 @@
----
-name: close-the-loop
-description: Process a call transcript into the full post-call loop. Files the transcript, drafts a personalized Gmail follow-up in the user's voice, logs commitments, appends a touchpoint to the contact's people file, and archives the source. The Gmail draft is NEVER auto-sent. Use when the user shares a transcript and wants the *whole* post-call workflow run, says "close the loop on this call", "process and draft followup for [person]", "wrap up the call with [person]".
-effort-level: medium
-invocation: explicit
-argument-hint: "[transcript or person name]"
----
-
-# Close the Loop
-
-The full post-call orchestration. One input (a transcript), four artifacts, one drafted external action that stays human.
-
-This skill is a *chain* on top of two existing skills:
-
-- It starts where `capture-meeting` ends: transcript filed, commitments extracted, participants identified.
-- It extends with the work `follow-up-draft` does: a personalized email built in the user's voice.
-- It adds the multi-write fanout: people file touchpoint, commitments ledger append, transcript archive.
-
-If the user only wants the capture half (no email drafted), route to `capture-meeting`.
-If the user only wants the email half (capture already happened), route to `follow-up-draft`.
-If they want the whole loop, this is the canonical skill.
-
-## When to fire
-
-Trigger phrases:
-
-- "Close the loop on this call"
-- "Process my call with [person] and draft the followup"
-- "Wrap up the call with [person]"
-- "I just got off a call with [person] — full workflow"
-- A transcript dropped into a designated `transcripts-inbox/` folder when the user has the close-the-loop pattern configured
-
-Do NOT fire for:
-
-- General document processing (use `file-document` or `ingest-sources`)
-- Pre-meeting prep (use `meeting-prep`)
-- Generic email response with no meeting context (use `draft-reply`)
-
-## The flow
-
-| Step | Action | Touches |
-|------|--------|---------|
-| 1 | File raw transcript via `memory_file` | source preservation |
-| 2 | Extract decisions, commitments (yours and theirs), unresolved questions | in-memory reasoning |
-| 3 | Read `people/{contact}.md` for relationship context | files |
-| 4 | Read recent Gmail history with the contact (tone calibration) | Gmail connector |
-| 5 | Read `voice-followup.md` if it exists | files |
-| 6 | Draft personalized email in Gmail Drafts | Gmail connector (DRAFT only, never send) |
-| 7 | Append dated touchpoint to `people/{contact}.md` | files |
-| 8 | Append commitments to `commitments.md` with deadlines | files |
-| 9 | Move transcript from `transcripts-inbox/` to `archive/transcripts/{contact}/` | files |
-| 10 | Report back: "Draft ready in Gmail. Logged N commitments. Touchpoint added to [contact]'s file." | chat |
-
-## The human gate (non-negotiable)
-
-The email **never sends automatically**. It lands in Gmail Drafts. The user opens, reviews, edits if needed, and hits Send themselves.
-
-This is not a limitation. It is the design. External communication is always Human-Approved. See `claudia-principles.md` § 1 (Safety First).
-
-If a future version of this skill ever auto-sends, that is a breaking change that requires an explicit logged decision.
-
-## Processing steps
-
-### 1. File the source (MANDATORY)
-
-Same as `capture-meeting`. Call `memory_file` with:
-
-- `filename`: `YYYY-MM-DD-[contact]-[topic].md`
-- `source_type`: `transcript`
-- `summary`: brief one-line summary
-- `about`: `[contact-slug]`
-- `content`: the full raw transcript
-
-### 2. Extract via Document Processor
-
-Dispatch the Document Processor agent (Haiku) with `extraction_type: "memory_operations"`. The agent returns commitments, decisions, and key topics as a structured operations array. Review for accuracy before storing.
-
-### 3. Read relationship and voice context
-
-In parallel:
-
-- `people/{contact}.md` (if exists; if not, plan a touchpoint entry that will create the file)
-- Recent Gmail thread with the contact (last 5 to 10 messages)
-- `voice-followup.md` from the user's playground or context folder
-
-If `voice-followup.md` does not exist, fall back to a neutral professional tone and flag to the user: "I drafted in a neutral voice. Create `voice-followup.md` to personalize."
-
-### 4. Draft the email
-
-Follow the structure from `voice-followup.md`. Default pattern when no voice file:
-
-- Opening: reference the call directly. No "I wanted to circle back."
-- Body beat 1: what *they* said that mattered (one or two sentences).
-- Body beat 2: what *you* committed to, with dates.
-- Closing: one clear next step with a specific date or window.
-- Signoff: "Best," then first name. No "Warm regards." No em dashes.
-
-Create the draft via the Gmail connector. Do NOT call any send tool. The draft must remain in Drafts.
-
-### 5. The multi-write fanout
-
-In sequence:
-
-- `memory_batch` with the reviewed memory operations from step 2.
-- Append to `people/{contact}.md` under the History section. Newest first. Date-stamped. Format: `### YYYY-MM-DD. [Topic].`
-- Append to `commitments.md` under "Active: things I owe" and "Active: things others owe me". Include source attribution: `*(source: call YYYY-MM-DD)*`.
-- Move the transcript file from `transcripts-inbox/` to `archive/transcripts/{contact-slug}/`.
-
-### 6. Report
-
-```
-Draft ready in Gmail. Open Drafts to review and send.
-Logged {N} commitments to commitments.md ({M} mine, {N-M} theirs).
-Touchpoint added to people/{contact}.md.
-Transcript archived to archive/transcripts/{contact-slug}/.
-```
-
-If anything failed (Gmail connector expired, person file write blocked, etc.), report the failure clearly and offer the recovery path. Do not silently proceed.
-
-## Failure modes
-
-| Failure | Recovery |
-|---------|----------|
-| Transcript has missing speaker labels | List commitments under "Active: unattributed" rather than guessing who owes what |
-| No `people/{contact}.md` file exists | Offer to create one with what was learned. Do not auto-create silently. |
-| `voice-followup.md` missing | Use neutral tone, flag the gap to the user |
-| Gmail connector expired or disconnected | Skip the draft step, log everything else, instruct the user to reconnect Gmail and re-run from step 4 |
-| Multiple contacts in the transcript | Ask the user which contact "owns" the followup before drafting. Do not draft to multiple recipients. |
-
-## See also
-
-- `capture-meeting` — the upstream half. Use when no email is needed.
-- `follow-up-draft` — the downstream half. Use when capture already happened.
-- `meeting-prep` — the inverse. Use before a call, not after.
-- `commitment-detector` — proactive sibling. Fires on commitments mentioned in conversation, separate from this skill which processes formal transcripts.
-
-## AIAC context
-
-This is the anchor workflow for the AIAC Cowork course (`~/Downloads/Claude-Cowork-Course/`). Module 3 of the course walks members through building this exact skill from the constituent capture-meeting and follow-up-draft pieces. The course's `voice-followup.md` template is the recommended companion file.

package/template-v2/.claude/skills/manuscript-fact-check/SKILL.md

@@ -1,111 +0,0 @@
----
-name: manuscript-fact-check
-description: Audit every quote and source claim in a draft manuscript against primary sources before it ships to a publisher, editor, or peer reviewer. Produces a structured per-claim matrix (claim, source verbatim, status, action, correction), applies corrections to the draft, and flags any remaining items that need physical-copy verification. Use when the user says "fact-check this", "verify the sources", "audit this manuscript", "check every quote", "build a fact-check matrix", "verify against primary sources", or before any publisher / editor submission.
-effort-level: high
-invocation: explicit
-argument-hint: "[path to draft or article]"
----
-
-# Manuscript Fact-Check
-
-The pass that runs before a draft ships. Every quote attributed to a source, every research claim, every statistic, verified against the primary source. Drift caught, corrections applied, audit trail preserved.
-
-This skill exists because Claudia's most catastrophic failure mode in writing work is the subtly-rewritten "quote" that drifts from the original. Editors at credible publishers catch these instantly. The chapter that ships with a misquote loses the editor's trust on page one, and the relationship rarely recovers.
-
-## When to fire
-
-Trigger phrases:
-
-- "Fact-check this manuscript / chapter / article / paper"
-- "Verify the sources in this draft"
-- "Audit every quote in [file]"
-- "Check the citations before I send this"
-- "Build a fact-check matrix for [draft]"
-- "Make sure these quotes are accurate against the originals"
-- "I'm sending this to [publisher]: verify everything first"
-
-Do NOT fire for:
-
-- Polishing prose, restructuring sections, or style edits (different skills)
-- Verifying a single fact in conversation (just look it up directly)
-- Drafts that contain no quoted sources or specific statistics
-
-## The three-tier strategy
-
-Before any verification begins, classify every claim by tier. The tier determines whether the claim is verified directly by Claudia or delegated to a research-scout agent.
-
-| Tier | What it is | Why direct verification | Examples |
-|------|------------|--------------------------|----------|
-| 1 | Quotes the publisher *owns* or that an audience knows verbatim | The publisher's editor will compare letter-by-letter. Any drift kills credibility. Never delegate. | Hill quotes for Napoleon Hill publisher. Bible quotes for theological work. Constitution quotes for legal work. The author's own prior books. |
-| 2 | Peer-reviewed research claims | Verifiable through DOI / PubMed / journal archive. Agents are good at this. | Neuroscience studies, meta-analyses, survey data, citation chains. |
-| 2.5 | Biographical quotes with retelling drift | Verifiable but commonly distorted in derivative sources. Agents work but must compare to primary source, not Wikipedia. | McCartney "Yesterday" story, Kekulé benzene dream, Jobs interviews, founder origin stories. |
-
-**Rule:** Tier 1 is verified by Claudia personally via `firecrawl scrape` + `grep` against the primary public-domain text. Tiers 2 and 2.5 are dispatched to parallel research-scout agents with strict "quote verbatim or flag UNVERIFIABLE" instructions.
-
-## The flow
-
-| Step | Action | Touches |
-|------|--------|---------|
-| 1 | Read the draft. Itemize every claim attributed to a source: direct quotes, paraphrased quotes, statistics, named studies, citations. | In-memory list. |
-| 2 | Classify each claim by tier (1, 2, 2.5). | In-memory list. |
-| 3 | For Tier 1 claims: identify the primary source (public-domain text, official transcript, the publisher's own catalog). Fetch directly with `firecrawl scrape`. Search the source with `grep` for each quoted phrase. Compare verbatim. | `sources/.firecrawl/` |
-| 4 | For Tier 2 and 2.5 claims: dispatch parallel research-scout agents. One agent for academic / peer-reviewed claims, one for biographical / example claims. Provide each agent with the exact manuscript wording and require structured output: claim, source verbatim, match status, confidence, recommended action. | Agent dispatch. |
-| 5 | Build the matrix. One row per claim. Columns: Claim (verbatim from manuscript), Source (verbatim from primary), Status (EXACT / PARAPHRASE / DRIFT / UNVERIFIABLE), Confidence (HIGH / MEDIUM / LOW), Action (KEEP / EDIT / FLAG), Notes. Save as `08-Fact-Check-Matrix.md` (or matching number) in the project bundle. | Writes file. |
-| 6 | Apply corrections to the manuscript source. For each DRIFT or PARAPHRASE flagged for EDIT: change the manuscript text to match the verified source wording. For each citation correction: update author names, years, page references. | Edits manuscript source. |
-| 7 | Re-render derived formats (PDF, DOCX, HTML) from the corrected source. | Re-runs the rendering pipeline. |
-| 8 | Report. Summarize what was verified EXACT, what required EDIT (with before / after), what remains FLAGGED for physical-copy verification, and what is now ready for the publisher. Honest about anything that could not be verified. | Chat output. |
-
-## Status codes
-
-| Code | Meaning |
-|------|---------|
-| **EXACT** | Manuscript wording matches the primary source verbatim. KEEP. |
-| **PARAPHRASE** | Manuscript captures the source meaning but rewords slightly. May be acceptable (with quotation marks removed) or require restoration of the verbatim quote. |
-| **DRIFT** | Manuscript presents text as a verbatim quote but the wording differs from the source. Always EDIT. |
-| **UNVERIFIABLE** | The source could not be reached or the quote could not be located within it. FLAG for the user to verify against a physical copy or alternate source. |
-
-## Output format (the matrix)
-
-Use this structure for each claim in `08-Fact-Check-Matrix.md`:
-
-```markdown
-### [Claim ID]. [Brief topic]
-
-| Field | Content |
-|-------|---------|
-| **Original claim** | "[verbatim from manuscript]" |
-| **Source (verbatim)** | "[verbatim from primary source]" |
-| **Status** | EXACT \| PARAPHRASE \| DRIFT \| UNVERIFIABLE |
-| **Action** | KEEP \| EDIT \| FLAG |
-| **Where it now stands** | "[corrected manuscript text, if action was EDIT]" |
-| **Source citation** | [author, title, publisher, year, page] |
-| **Notes** | [anything else the editor should know] |
-```
-
-End the matrix with a summary table and an "Outstanding items requiring action before submission" section that names every UNVERIFIABLE or FLAG claim and what the user must do to resolve it.
-
-## Why this skill matters
-
-For a publisher-facing submission, the editor's first read is a credibility test. A single fabricated quote, a wrong author initial, a confused year, a phrase attributed to one source when it comes from another: any of these tells the editor the rest of the manuscript cannot be trusted. The matrix prevents that. It also gives the editor a verification artifact: when they ask "where did this quote come from?", the answer is in the matrix, with the primary source cited and the verbatim text shown.
-
-For the writer, the matrix is the audit trail that turns "I think the quote is right" into "I can prove the quote is right."
-
-## Known failure modes
-
-| Failure | What goes wrong | How to handle |
-|---------|----------------|---------------|
-| Upstream research agents fabricate citations | Agents have produced false author initials, false dates, false page numbers when their search returns weak hits | Always have the agent quote the source verbatim, not paraphrase it. If the verbatim text cannot be produced, status is UNVERIFIABLE, not EXACT. |
-| The primary source is paywalled or behind anti-bot | WebFetch returns 403; agent reports the page as unverifiable | Try `firecrawl scrape` first. If still blocked, dispatch a different agent for that specific source. If still no, flag UNVERIFIABLE and ask the user to verify against a physical copy. |
-| The quote is widely attributed but the page reference is wrong | The quote is real, the page number is folklore | Mark the quote EXACT and the page number FLAG. The author must verify page references against a physical copy before final submission. |
-| The quote is "from" a famous person but only sourced to a later retelling | Common with Tesla, Edison, Twain, Einstein quotes | Skip the quote if the primary source is the person's own late autobiography or a posthumous secondary citation. Note exclusion in the editor's notes. |
-
-## AIAC context
-
-The reference implementation of this skill is the Napoleon Hill "Sixth Sense" chapter fact-check at `~/Downloads/Napoleon-Hill-Sixth-Sense/08-Fact-Check-Matrix.md`. 26 claims audited, ~14 EDITs applied, two FLAGs preserved for physical-copy verification. The chapter was meaningfully strengthened by the pass: at least three errors that would have been caught by a serious editor were caught here first.
-
-## See also
-
-- `auto-research` for iterative draft improvement (different concept: this audits sources, that iterates on prose quality)
-- `summarize-doc` for executive summaries (different concept: this verifies an existing draft, that produces a new artifact)
-- `ingest-sources` for multi-document processing (related concept: that processes raw inputs, this audits already-cited outputs)
-- `capture-meeting` for meeting transcript processing (related concept for transcript-source claims, but different output)