fablever

For genuinely hard multi-part work, write a durable decision-complete plans/{slug}.md before execution, then build against it. Use when a task has roughly four or more distinct parts, spans several files, or has real ordering/dependency. Pulled only by an explicit trigger; never a default/always-on stage.

Apply the deterministic route-vs-solo cost preflight to an opted-in orchestration task. Default to one agent with the fixed lens menu; launch a Workflow recipe only when its multi-agent cost floor allows it. Pulled on demand; not always-on.

For a genuinely hard, multi-part task, externalize a short numbered plan BEFORE executing, then build against it — so the deliverable is better organized and more correct than diving straight in. Use when a task has roughly four or more distinct parts, spans several files, or has ordering/dependency between steps. Pulled on demand; not always-on, and a no-op on simple or single-step work.

Run an acceptance check before handing over an external-facing deliverable — research summaries, docs, marketing/funnel copy, a report, a code-change summary, anything a person will act on. Use right before you present a "final" artifact, or when the user says "send it", "ship it", "is this ready", "finalize", "draft the deliverable". Checks acceptance criteria, not vibes. Pulled on demand; not always-on.

Before you say "done", "fixed", "works", "passing", or "verified", ground that claim in a real command/test/file/tool result on the same line — or say plainly that it is not verified yet. Use whenever you are about to report completion or success of a code change, a build, a fix, or a task. Catches unsupported "it works" claims. Pulled on demand; not always-on.

Adversarially review a change or artifact before it is locked in — try to break it, find the failure paths, rate severity — instead of confirming it looks fine. Use before merging/shipping/finalizing, or when the user says "review this", "poke holes", "what breaks", "find the weaknesses", "red-team it", "what did I miss". Finds and rates defects; does not rewrite. Pulled on demand; not always-on.

Hold the exact scope the user asked for — no extra files, no refactors, no "while I'm here" cleanup, no speculative features. Use when the user says "only", "just", "do not edit", "minimal change", "report only", "no refactor", "don't touch X", or otherwise draws a narrow boundary; and as a check before you touch any file the request did not name. Pulled on demand; not always-on.

Generate or refresh a local AGENTS.md for a module by reading the code that already exists, so later work follows the project's real conventions instead of model defaults. Use when starting work in an unfamiliar module, when conventions keep getting violated, or when asked to "seed", "init", or "write AGENTS.md for" a directory. Pulled on demand; not always-on.

Open a long or multi-session report with a fixed 3-line [Handoff Summary] (context / what changed / the single decision to act on), and keep a per-session .fablever_state file so an operator juggling many projects can reload context in one glance. Use when finishing a long or multi-step task, reporting back into a project the operator has been away from, or asked to "hand off", "update the state file", or "write the handoff". On-demand; not always-on, and a no-op on short or conversational turns.

Generate or refresh local AGENTS.md convention files for a codebase by reading the code that already exists, so later work follows the project's real conventions instead of model defaults. Use when starting work in an unfamiliar module, when conventions keep getting violated, or when asked to "seed", "init", or "write AGENTS.md for" a directory. Pulled on demand; not always-on.

Route a task to a proven multi-agent Workflow recipe (independent adversarial review, divergent exploration, decompose-and-fan-out, staged map, or best-of-N judge panel) instead of answering inline. Use when a task is genuinely multi-part, needs independent verification, needs many distinct ideas, or processes many items — and the user has opted into multi-agent orchestration. Pulled on demand; not always-on.