Think-Short

Toggle terse-output mode. Writes state to $CLAUDE_PLUGIN_DATA/think-short.json (global) or .claude/brewtools/think-short.json (project). Hooks (authored separately) read state and inject profile-specific directives into SessionStart + PreToolUse:Task. This skill ONLY parses intent and mutates state.

Robustness Rules

BT_ROOT Resolver

$CLAUDE_PLUGIN_ROOT is NOT inherited by the Bash tool in main-conversation slash invocations. Every Bash block MUST resolve BT_ROOT dynamically. Use the env var if present, else pick the newest cached version (survives version bumps — no hardcoded 3.7.4):

BT_ROOT="${CLAUDE_PLUGIN_ROOT:-$(ls -d ~/.claude/plugins/cache/claude-brewcode/brewtools/*/ 2>/dev/null | sort -V | tail -1 | sed 's:/*$::')}"
test -d "$BT_ROOT/skills/think-short/helpers" || { echo "❌ BT_ROOT invalid: $BT_ROOT"; exit 1; }

Paths (use $BT_ROOT literally in Bash):

• Global state: $CLAUDE_PLUGIN_DATA/think-short.json (fallback: ~/.claude/plugins/data/brewtools-claude-brewcode/think-short.json) — computed by getPaths(cwd)
• Project state: $PWD/.claude/brewtools/think-short.json — computed by getPaths(cwd)
• State helper: $BT_ROOT/skills/think-short/helpers/state.mjs — exports getPaths, readPluginDefaults, resolveEffectiveState, writeState, log
• Safe-write helper: $BT_ROOT/skills/think-short/helpers/safe-write.mjs — exports safeReadJson, safeWriteJson
• Log file: $PWD/.claude/logs/brewtools.log (auto-created by log())

State schema:

{"version":1, "enabled":false, "profile":"medium", "blacklist":["debate","docs-writer","architect"], "updated_at":"ISO"}

P0: Parse Intent

Parse $ARGUMENTS into structured form:

{ op: on|off|profile|status|blacklist, profile?: light|medium|aggressive, blacklistOp?: add|remove, agent?: string, scope?: global|project }

Structural match (exact)

NL-prompt fallback (MANDATORY)

If no structural match, treat argument as NL prompt. Algorithm:

1. Trim + lowercase.
2. Tokenize + apply synonym table:

3. Combos allowed — e.g. включись максимально → on + profile aggressive. Execute BOTH ops in sequence.
4. Ambiguous (0 matches OR >1 mutually-exclusive match that isn't a combo) → AskUserQuestion with candidate operations as options.
5. After resolution, INFO log via utils.mjs:

   think-short: NL-prompt "<input>" → resolved as <command>
   

EXECUTE using Bash tool (resolve + log):

BT_ROOT="${CLAUDE_PLUGIN_ROOT:-$(ls -d ~/.claude/plugins/cache/claude-brewcode/brewtools/*/ 2>/dev/null | sort -V | tail -1 | sed 's:/*$::')}"
test -d "$BT_ROOT/skills/think-short/helpers" || { echo "❌ BT_ROOT invalid: $BT_ROOT"; exit 1; }
node --input-type=module -e "
import {log} from '${BT_ROOT}/skills/think-short/helpers/state.mjs';
log('info', 'NL-prompt \"INPUT\" → resolved as RESOLVED', process.cwd(), process.env.CLAUDE_CODE_SESSION_ID || null);
" && echo "OK log" || echo "FAILED log"

Replace INPUT and RESOLVED literally. The log() from state.mjs auto-prefixes with think-short — do NOT add it again.

P1: Scope Selection

Default = project scope. Silent — no AskUserQuestion unless the user explicitly asks for disambiguation.

Rules:

ALWAYS log the chosen scope at INFO using log() from state.mjs:

think-short: scope=<project|global> (<default|--scope|user-choice>, --scope <not specified|explicit>)

For status — no scope question (reads merged state). For blacklist — defaults to project scope silently.

If user is invoking from a combo (e.g. включись максимально) — determine scope ONCE via the rules above, apply to all ops in the combo.

P2: Mutate State

EXECUTE using Bash tool (generic pattern — substitute SCOPE, PATCH_JSON, OP):

BT_ROOT="${CLAUDE_PLUGIN_ROOT:-$(ls -d ~/.claude/plugins/cache/claude-brewcode/brewtools/*/ 2>/dev/null | sort -V | tail -1 | sed 's:/*$::')}"
test -d "$BT_ROOT/skills/think-short/helpers" || { echo "❌ BT_ROOT invalid: $BT_ROOT"; exit 1; }
node --input-type=module -e "
import {writeState, log} from '${BT_ROOT}/skills/think-short/helpers/state.mjs';
const patch = PATCH_JSON;
const r = await writeState('SCOPE', patch, process.cwd());
log('info', 'toggle OP applied (scope=SCOPE) → ' + JSON.stringify(patch), process.cwd(), process.env.CLAUDE_CODE_SESSION_ID || null);
console.log(JSON.stringify({scope:'SCOPE', file:r.path, state:r.after}));
" && echo "OK mutate" || echo "FAILED mutate"

Substitutions per op:

writeState handles: reading existing scope file, merging defaults, atomic write via safeWriteJson, stamping updated_at, enforcing version:1. No manual fs.existsSync / safeWrite calls needed.

Combo ops (e.g. on + profile aggressive): pass a single merged patch: {enabled:true, profile:'aggressive'} — one writeState call, atomic.

Blacklist mutation example (inline — single node invocation reads current state then writes):

BT_ROOT="${CLAUDE_PLUGIN_ROOT:-$(ls -d ~/.claude/plugins/cache/claude-brewcode/brewtools/*/ 2>/dev/null | sort -V | tail -1 | sed 's:/*$::')}"
node --input-type=module -e "
import {resolveEffectiveState, writeState, log} from '${BT_ROOT}/skills/think-short/helpers/state.mjs';
const s = await resolveEffectiveState(process.cwd());
const cur = Array.isArray(s.blacklist) ? s.blacklist : [];
const next = Array.from(new Set([...cur, 'AGENT'