feat: show Claude Code accounts in list and forecast

docs/superpowers/specs/2026-05-19-skill-profile-picker-design.md

@@ -0,0 +1,230 @@
+# Skill-Profile Picker on Agent Launch
+
+**Status:** Draft
+**Date:** 2026-05-19
+**Author:** webubusiness@gmail.com (NagyVikt)
+**Follows:** PR #35 (Specialize authmux launches by skill profile), PR #36 (hermes-agent third target)
+
+## Problem
+
+PR #35 + PR #36 wire authmux to silently activate a per-account Soul skill profile
+before launching `codex`, `claude-account1`, `claude-account2`, etc. The profile
+is either fixed by `AUTHMUX_DEFAULT_SKILL_PROFILE`, by the account's saved
+`.authmux-skill-profile` file, or hard-coded `base`.
+
+That is too rigid. Every terminal session has a different *task shape* —
+frontend redesign, deploy, code review, orchestration. The user wants to pick
+the skill profile **at launch time**, the same way they pick an account, instead
+of editing files or remembering a CLI flag.
+
+## Goal
+
+When the user types `codex`, `claude-account1`, or `claude-account2` in a fresh
+terminal, the existing shell hook prompts them with an interactive arrow-key
+list of Soul skill profiles (frontend, design, medusa, review, orchestration,
+deploy, base, all, …). They press Enter; the selected profile is activated
+against the right target directory and saved as the new default for that
+account; then the underlying agent launches normally.
+
+Non-interactive shells (CI, pipes, `--quiet`) skip the prompt and silently
+fall back to the saved/default profile — current behavior is preserved.
+
+## Non-goals
+
+- **No new profile content.** Authmux still consumes `~/Documents/soul/skills/profiles/*.json`. Adding profiles is a soul-repo concern.
+- **No MCP bundling.** Same scope boundary PR #35 took: MCP server selection stays in soul-side installers (`install-codex-mcps.py`, `install-claude-mcps.py`). Out of scope here.
+- **No global "session manager" UI.** This is one prompt at launch, not a TUI.
+- **No new hermes wrapper.** `hermes` does not have a shell-launch hook today; it remains opt-in via `authmux skills activate --agent hermes`.
+
+## Success criteria
+
+1. Running `claude-account2` in an interactive terminal shows a select prompt seeded with the account's currently-saved profile, lets the user change it with arrow keys, and launches `claude` against `~/.claude-accounts/account2` with the chosen skills directory rebuilt.
+2. Running `codex` in an interactive terminal does the same against `~/.codex/skills`.
+3. Pressing Ctrl-C at the prompt aborts cleanly without launching the agent (so user can re-type).
+4. The chosen profile persists in `<account-dir>/.authmux-skill-profile` so a second invocation defaults to the previous choice.
+5. Setting `AUTHMUX_SKILL_PICK=off` (or piping non-TTY input) restores the silent PR #35 behavior.
+6. Setting `AUTHMUX_SKILL_PROFILE=<name>` for one command bypasses the prompt and uses the override.
+7. Existing tests still pass; new tests cover the non-TTY fallback and the env-bypass path.
+
+## Architecture
+
+### One new command: `authmux skills pick`
+
+A thin command that wraps the existing `listAvailableSkillProfiles()` + `activateSkillProfile()` building blocks with an interactive `prompts` select. Lives in `src/commands/skills-pick.ts` (separate file rather than overloading `skills.ts` so its presence in the hook is grep-able and unit-testable).
+
+Flags:
+
+| Flag | Meaning |
+|------|---------|
+| `--agent <codex\|claude\|hermes>` | Agent target (same as `skills activate`). |
+| `--target <dir>` | Explicit skills directory (skips `resolveDefaultSkillTarget`). |
+| `--account <name>` | Save the chosen profile to this account; defaults to current authmux account. |
+| `--save` / `--no-save` | Persist the choice (default: `--save`). For `--agent claude` with `--account`, writes `~/.claude-accounts/<account>/.authmux-skill-profile`. For `--agent codex` (no `--account` flag), resolves to `accounts.getCurrentAccountName()` and writes via `accounts.setSkillProfileForAccount(name, profile)` — same path the existing `skills use` command takes. If no current account is set, `--save` becomes a no-op with a warning. |
+| `--default <name>` | Profile to pre-select / fall back to. Default cascade: `AUTHMUX_SKILL_PROFILE` → account saved → `AUTHMUX_DEFAULT_SKILL_PROFILE` → `base`. |
+| `--no-activate` | Save-only; skip running the activator. |
+| `--quiet` | Suppress the activation summary line; errors still print. |
+| `--json` | Standard JSON envelope (Theme X4). |
+
+Behavior:
+
+1. Resolve the default profile via the cascade.
+2. Resolve the target directory: explicit `--target` wins; else `resolveDefaultSkillTarget(agent)`; else for claude with `--account`, `~/.claude-accounts/<account>/skills`.
+3. Decide interaction mode:
+    - `AUTHMUX_SKILL_PICK=off` → skip prompt, use default.
+    - `--json` → skip prompt (corrupts stdout), use default.
+    - `!process.stdin.isTTY || !process.stdout.isTTY` → skip prompt, use default.
+    - `AUTHMUX_SKILL_PROFILE` set → skip prompt, use that value.
+    - Otherwise → show prompt.
+4. If prompting, show a `prompts.select` with profile names from `listAvailableSkillProfiles()`, `initial` = index of default. On Ctrl-C, exit code 130, no activation, no save.
+5. Activate via `activateSkillProfile({ profile, agent, target })`.
+6. If `--save` and `--account`/current account is known, write `<account-dir>/.authmux-skill-profile`.
+7. Print one summary line: `Skill profile: <name> → <target> (<n> skills)`.
+
+### Hook rewrites
+
+#### Codex wrapper (`scripts/postinstall-login-hook.cjs`)
+
+Inside the generated `codex()` shell function, replace the silent
+
+```
+command authmux skills activate-current --agent codex >/dev/null 2>&1 || true
+```
+
+with an interactive call that keeps stdin/stdout attached:
+
+```
+command authmux skills pick --agent codex --quiet --save || true
+```
+
+Trailing `|| true` preserves the launch-anyway contract. We do **not** swallow stderr — pick errors should be visible (missing activator, soul not installed). `--quiet` suppresses the success line so the prompt feels lightweight.
+
+#### Claude parallel aliases (`src/commands/parallel.ts` → `generateAliases()`)
+
+Today's alias is one line that activates silently, then launches claude. Move the activation off the alias and into a small shell function `__authmux_claude_account()` emitted once at the top of the managed block, then make each alias call that function with the account name. The function:
+
+```
+__authmux_claude_account() {
+  local name="$1"
+  local dir="$HOME/.claude-accounts/$name"
+  command authmux skills pick \
+    --agent claude \
+    --account "$name" \
+    --target "$dir/skills" \
+    --quiet \
+    --save || true
+  CLAUDE_CONFIG_DIR="$dir" command claude
+}
+alias claude-account1='__authmux_claude_account account1'
+alias claude-account2='__authmux_claude_account account2'
+```
+
+Crucially, no `>/dev/null` redirection on the `pick` call — the prompt needs the terminal. The function preserves the existing exit-code semantics (the alias still exits with claude's exit code).
+
+#### Env knobs
+
+Documented in README and accepted by `pick`:
+
+| Variable | Effect |
+|----------|--------|
+| `AUTHMUX_SKILL_PICK=off` | Always silent fallback (restores PR #35 behavior). |
+| `AUTHMUX_SKILL_PROFILE=<name>` | One-shot override, no prompt. |
+| `AUTHMUX_DEFAULT_SKILL_PROFILE=<name>` | Fallback when no account-saved profile (already existed). |
+
+### Files touched
+
+```
+src/commands/skills-pick.ts        (new)
+src/lib/skills/profile.ts          (small additions: profile-cascade helper, isInteractive())
+src/tests/skills-pick.test.ts      (new)
+src/tests/skills-profile.test.ts   (extend: cascade helper)
+src/commands/parallel.ts           (generateAliases — emit function + aliases)
+scripts/postinstall-login-hook.cjs (codex() body — swap activate-current → pick)
+README.md                          (one section + env table additions)
+```
+
+### Data flow
+
+```
+user types `claude-account2`
+        │
+        ▼
+__authmux_claude_account account2
+        │
+        ▼
+authmux skills pick --agent claude --account account2 --target ~/.claude-accounts/account2/skills
+        │
+        ├─ TTY? AUTHMUX_SKILL_PICK=off? AUTHMUX_SKILL_PROFILE set? --json?
+        │       │
+        │       └─ silent path → use saved/default profile
+        │
+        ▼
+prompts.select { profiles, initial = saved-or-default }
+        │
+        ▼
+activateSkillProfile(profile, agent=claude, target)
+        │
+        ▼
+write <account-dir>/.authmux-skill-profile (--save default)
+        │
+        ▼
+CLAUDE_CONFIG_DIR=… command claude
+```
+
+### Error handling
+
+| Situation | Behavior |
+|-----------|----------|
+| Activator script missing | `pick` warns, returns activated=false, agent still launches (preserve PR #35 contract). |
+| Activator script fails | `pick` prints stderr; `|| true` in alias means agent still launches. |
+| User Ctrl-C at prompt | `pick` exits 130, no save, no activate; `|| true` in alias means agent still launches with *previously* activated skills (no rebuild — safe). |
+| No profiles found (soul missing) | `pick` prints "no skill profiles available — set AUTHMUX_SOUL_HOME", returns activated=false, agent still launches. |
+| Profile name invalid | `normalizeSkillProfileName` throws; pick prints error; agent still launches. |
+
+## Testing
+
+Unit tests (`src/tests/skills-pick.test.ts`):
+
+1. **Non-TTY fallback:** with `process.stdin.isTTY = false`, `pick --agent codex` activates the default profile without prompting; verify `activateSkillProfile` call signature.
+2. **Env bypass:** with `AUTHMUX_SKILL_PROFILE=frontend`, pick uses `frontend` regardless of saved profile.
+3. **Save default:** non-TTY run with `--save --account testacc` writes `<dir>/.authmux-skill-profile`.
+4. **No-save:** `--no-save` skips the write.
+5. **Cascade order:** new `resolveDefaultProfileFor(agent, accountName)` helper returns env > account-saved > AUTHMUX_DEFAULT_SKILL_PROFILE > "base".
+6. **Activator missing:** stub a missing activator path, assert `activated=false` with a warn line, exit code 0.
+
+Integration smoke (manual, documented in spec; no harness yet):
+
+- Start a fresh terminal, type `claude-account2`, pick `design`, confirm `~/.claude-accounts/account2/skills` contains the design profile's symlinks.
+- Type `claude-account2` again, confirm `design` is pre-selected.
+
+## Parallel implementation tracks
+
+The change naturally splits into three tracks that share no files. They should be implemented by parallel sub-agents:
+
+| Track | Owns | Touches |
+|-------|------|---------|
+| **A — TS core** | New command + helper + types | `src/commands/skills-pick.ts`, `src/lib/skills/profile.ts`, `src/index.ts` (oclif registration if needed) |
+| **B — Shell hooks** | Wire pick into codex + claude wrappers | `scripts/postinstall-login-hook.cjs`, `src/commands/parallel.ts` (only `generateAliases()` / `installAliases()` blocks) |
+| **C — Tests + docs** | Coverage + user-facing docs | `src/tests/skills-pick.test.ts`, `src/tests/skills-profile.test.ts`, `README.md`, this spec |
+
+Tracks A and B both depend on the spec being final; B depends on A's flag surface being agreed (the spec freezes it). Track C depends on A for tests but can write the README skeleton in parallel.
+
+Sequencing for the parallel run:
+
+1. Spec merges first (this doc).
+2. Tracks A + C-docs start in parallel.
+3. Track B starts once A's flag names are committed (~1 commit in).
+4. Track C-tests start once A's helper signatures land.
+5. All three rebase onto a single integration branch, then one PR.
+
+## Open questions (resolved by author for V1)
+
+- **Q: Auto-timeout the prompt?** No. If the user wants no prompt, they set `AUTHMUX_SKILL_PICK=off`. Auto-timeout adds non-determinism for one shaved keystroke.
+- **Q: Show a per-profile description in the prompt?** Not in V1. Profile JSON has no description field. If users want this, add it to soul-side JSON first, then wire it through.
+- **Q: Cache the profile list?** No. `listAvailableSkillProfiles()` is a single `readdirSync` on ~10 entries — sub-millisecond.
+- **Q: Should `authmux use <account>` also prompt?** Not in V1. `use` is non-interactive by default and is often scripted. The prompt belongs in the launch wrapper, not the account switcher.
+
+## Migration / backwards compat
+
+- Users on PR #35 get the new behavior automatically the next time they reinstall (`npm install` triggers `postinstall-login-hook.cjs`) **and** the next time they run `authmux parallel --install`. Until they do, the old aliases keep silent-activating; nothing breaks.
+- The new `pick` command is additive. `skills activate-current` is unchanged; the only callsite that switched is the shell hook block.
+- `AUTHMUX_SKILL_PICK=off` is the escape hatch for users who liked the silent flow.

src/commands/forecast.ts

@@ -1,6 +1,7 @@
 import { BaseCommand } from "../lib/base-command";
 import { AccountService } from "../lib/accounts/account-service.js";
 import { forecastAccounts } from "../lib/account-health.js";
+import { listClaudeAccounts } from "../lib/accounts/claude-health";
 
 export default class Forecast extends BaseCommand {
   static description = "Show health forecast for all saved accounts (best-first)";
@@ -21,19 +22,35 @@ export default class Forecast extends BaseCommand {
       const service = new AccountService();
       const names = await service.listAccountNames();
       const forecasts = names.length ? forecastAccounts(names) : [];
+      const claudeAccounts = listClaudeAccounts();
 
-      this.emit({ accounts: forecasts }, (data) => {
-        if (!data.accounts.length) {
+      this.emit({ accounts: forecasts, claude: claudeAccounts }, (data) => {
+        if (!data.accounts.length && !data.claude.length) {
           this.log("No saved accounts found.");
           return;
         }
-        this.log("Account Health Forecast (best first):\n");
-        for (let i = 0; i < data.accounts.length; i++) {
-          const h = data.accounts[i];
-          const status = h.usable ? "✓" : "✗";
-          this.log(
-            `  [${i + 1}] ${status} ${h.name}: score=${Math.round(h.score)} circuit=${h.circuitState} tokens=${Math.round(h.tokensAvailable)}`,
-          );
+        if (data.accounts.length) {
+          this.log("Codex Account Health Forecast (best first):\n");
+          for (let i = 0; i < data.accounts.length; i++) {
+            const h = data.accounts[i];
+            const status = h.usable ? "✓" : "✗";
+            this.log(
+              `  [${i + 1}] ${status} ${h.name}: score=${Math.round(h.score)} circuit=${h.circuitState} tokens=${Math.round(h.tokensAvailable)}`,
+            );
+          }
+        }
+        if (data.claude.length) {
+          this.log("\nClaude Code Account Health:\n");
+          for (let i = 0; i < data.claude.length; i++) {
+            const ca = data.claude[i];
+            const status = ca.healthy ? "✓" : "✗";
+            const daysLeft = ca.expiresAt
+              ? Math.max(0, Math.round((ca.expiresAt - Date.now()) / 86400000))
+              : 0;
+            this.log(
+              `  [${i + 1}] ${status} ${ca.name}: type=${ca.subscriptionType} expires_in=${daysLeft}d healthy=${ca.healthy}`,
+            );
+          }
         }
       });
     });

src/commands/list.ts

@@ -2,6 +2,7 @@ import { Flags } from "@oclif/core";
 import prompts from "prompts";
 import { BaseCommand } from "../lib/base-command";
 import { formatAccountType } from "../lib/accounts/plan-display";
+import { listClaudeAccounts } from "../lib/accounts/claude-health";
 import {
   fetchLatestNpmVersionCached,
   formatGlobalInstallCommand,
@@ -36,10 +37,11 @@ export default class ListCommand extends BaseCommand {
       }
 
       const accounts = await this.accounts.listAccountMappings({ refreshUsage: "missing" });
+      const claudeAccounts = listClaudeAccounts();
 
-      this.emit({ accounts, detailed }, (payload) => {
-        if (!payload.accounts.length) {
-          this.log("No saved Codex accounts yet. Run `authmux save <name>`.");
+      this.emit({ accounts, claudeAccounts, detailed }, (payload) => {
+        if (!payload.accounts.length && !payload.claudeAccounts.length) {
+          this.log("No saved accounts yet. Run `authmux save <name>` or `authmux parallel --add <name>`.");
           return;
         }
 
@@ -59,6 +61,20 @@ export default class ListCommand extends BaseCommand {
             `    type=${formatAccountType(account.planType)} plan=${account.planType ?? "-"} skillProfile=${account.skillProfile ?? "-"} usage=${account.usageSource ?? "-"} 5h=${this.formatRemaining(account.remaining5hPercent)} weekly=${this.formatRemaining(account.remainingWeeklyPercent)} lastUsageAt=${account.lastUsageAt ?? "-"}`,
           );
         }
+
+        if (payload.claudeAccounts.length > 0) {
+          if (payload.accounts.length > 0) this.log("");
+          this.log("Claude Code accounts:");
+          for (const ca of payload.claudeAccounts) {
+            const status = ca.healthy ? "✓" : "✗";
+            const expiry = ca.expiresAt
+              ? `expires=${new Date(ca.expiresAt).toISOString().slice(0, 10)}`
+              : "no-token";
+            this.log(
+              `  ${status} ${ca.name}  type=${ca.subscriptionType}  ${expiry}`,
+            );
+          }
+        }
       });
     });
   }

src/commands/skills.ts

@@ -32,7 +32,7 @@ export default class SkillsCommand extends BaseCommand {
     }),
     agent: Flags.string({
       description: "Agent skill target",
-      options: ["codex", "claude"],
+      options: ["codex", "claude", "hermes"],
       default: "codex",
     }),
     target: Flags.string({