agent-workbenchlisted

Use Agent Workbench as the project-local source of current accepted design, focused work, user corrections, KPT learning, Command Profiles, review findings, Lean proof receipts, evidence freshness, and completion readiness. Trigger when starting, resuming, handing off, reviewing, verifying, or completing coding-agent work in a project that uses `.agent-workbench`.
MuNeNiCK/agent-workbench · ★ 0 · AI & Automation · score 56

Install: claude install-skill MuNeNiCK/agent-workbench

# Agent Workbench Use the native Workbench executable for state and decisions. Do not reconstruct its ledger rules in the Skill or shell. ## Enter a project 1. Find the project root. 2. If `.agent-workbench/bin/agent-workbench` is absent, run `scripts/setup.sh PROJECT_ROOT` on POSIX or `scripts/setup.ps1 -ProjectRoot PROJECT_ROOT` on Windows. 3. Run `.agent-workbench/bin/agent-workbench --project PROJECT_ROOT context`. 4. Run `... describe` and select only from `applicableOperations`. Before using an unfamiliar mutation, run `... describe OPERATION`; require `applicable: true` and use only its `inputExample` fields. 5. Use only the returned current context for the next action. Query history separately when needed; do not treat matching old text as current. Current Context returns bounded stable references. Retrieve details with `design get`, `work get`, or `entry get`; page older entries with `history`. Do not replace the bounded context with a full ledger dump. The setup script only acquires and verifies the release archive. After setup, invoke the native binary directly. ## Operate safely The JSON on stdin is a command-specific machine transport, not a persisted record format. Never construct a `ProjectState` or `LedgerEntry`, inspect source to infer fields, or add fields absent from the native contract. Unknown fields and inapplicable operations must fail without a state revision change. ```text describe [operation] design propose | accept | get work st