AI Agent Workflow Collection

Read N recent worker tmux scrollbacks / handoff events, identify recurring friction, write a new skill or update an existing one.

Sole writer of ledger rows. Validates kind/type/state, mints slug ids, emits 'created' event. Adds decompose verb that turns a PRD into a chain of TDD slices.

TDD template for bookie subagent tasks. Covers pure-validator TDD, fixture patterns, table-driven cases, and the red-green-refactor loop specific to ledger write validation.

Hygiene phase skill. Fix doc/code drift, broken refs, stale ADRs, missing glossary terms, failed example commands. Docs-only PR, ≤3 files, ≤100 lines diff. Run via bookie create --kind task --class hygiene --title 'clarify-docs: '.

Headless-shaped invocation of claude that runs inside an interactive tmux pane you can attach to live. Gives you the shape of claude -p --output-format json with full transparency: every turn is observable in real time, attachable mid-flight, and inspectable after the fact via the session log.

Build Docker images and launch a local preview environment with docker-compose. Use to test the full stack locally before merging.

Slice-bounded refactor: clarify boundaries, remove incidental complexity, extract duplication. No behavior change.

Install a UserPromptSubmit + Stop hook pair that reminds the model to avoid sycophancy without bloating the system prompt.

Write a new knowledge entry to ~/vault/ke/ / and update the FTS5 index. One file per insight. Stop-hook on dev/director sessions.

FTS5 search over ~/vault/ke/ knowledge entries. Returns top-N markdown excerpts ranked by BM25, with file paths and headings.

Capture a real execution of a worker-shaped system, replay it against a candidate config in an isolated sandbox, diff the two. Use before promoting prompt changes, model swaps, skill set changes, or API upgrades. System-agnostic — works for any unit of work that has identifiable inputs, environment state, an observable transcript, and a measurable output.

Re-entrant sprint supervisor loop. Drives a single thin vertical slice to evidence-backed done across re-entry cycles: load prior handoffs, check evidence, complete or decompose, write re-entry handoff, tear down.

Owner-facing flow for filing a single ledger row. Walks the user through the required flags, then delegates the write to bookie.

Move files unused for ≥30d (or unreferenced per grep) into ~/trash/ - / with a TTL marker. Reversible; the actual sweep is run by Slice C machinery.

Agent-actor triage loop. Batch-claims rows with agent='agentunset' OR pool='poolunset', assigns {tier, pool, agent} per payload + context, writes via bookie.

Director-subagent skill. Classifies a state=failed ledger row as low-risk (auto-decompose into slices, cancel parent) vs needs-HITL (mark for human review).