agentsclimarketplace

Plan

Skill boshu2/agentops/images/gemini/skills/plan

The operating loop a coding agent follows — and skills to orchestrate multi-agent systems.

Install
npx -y skills add boshu2/agentops --skill plan

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Shape or refine the existing bead or caller intent without creating a second planning artifact. Triggers: "plan", "discover and plan", "shape this goal".

SKILL.md

5.2 KB, as published. Nobody here has run it

Plan

Turn the caller's intent into one bounded, testable behavior in the place that already owns the work. Prefer the caller's tracker, if any. When no tracker is available, use the caller's conversation or supplied issue text; the runtime snapshots the resolved intent bytes automatically so later contexts can read and hash the same source. Do not make the model restate those facts in a packet.

Workflow

  1. Resolve the intent source and choose one active behavior. When that source is not already durable, have the runtime pass its exact bytes to python3 skills/validate/scripts/validate.py snapshot-intent --source - and use the returned intent_ref for later phases.
  2. Route the work by type (see Ground-truth routing) and name its ground truth first. Then inspect only enough real context to make paths, interfaces, and evidence concrete. Existing research and specialist skills are advisory inputs.
  3. Ensure the source contains acceptance examples, important non-goals, and the allowed write scope. Use lightweight prose or Given/When/Then only where it removes ambiguity; do not require both normal and edge ceremony for every change.
  4. Name the first useful acceptance check.
  5. If authorized and the source is writable, update that bead or issue in place. Otherwise return a concise proposed amendment to the caller.

Planning produces no AgentOps packet. The runtime stores and hashes the resolved source bytes to detect later acceptance drift. That content-addressed snapshot is derived automatically and is not another model-authored planning artifact.

Bound the work around the caller-visible outcome, not individual files, gates, or reviewer comments. Decomposition is useful only when it reduces reasoning cost; it must not multiply invocations or proof artifacts.

Scope admission

In a repository with generated projections, write scope names generator-owned outputs as a class — the hand-edited sources plus all outputs of the owning regen commands — never as a hand-enumerated path list. Hand enumeration is falsified the first time a regen command rewrites a companion the author did not list: the 2026-07-15 heal-skill fold burned two implement lanes and three intent revisions (.agents/ao/intents/sha256/d1db59d4...2b81 superseded by f5fd7c3c...af75 superseded by 26a4f2be...eb48) before scope was restated as a class.

Before freezing acceptance, run a complexity admission: enumerate the generated companions, parity twins (for example a skills-codex/ mirror), and test files that assert on the paths being changed. Anything this pass finds that the scope does not admit will surface later as an out-of-scope diff or a broken gate.

Ground-truth routing

Every plan needs a ground truth outside the planner's own reasoning. Before freezing acceptance, classify the work and name its ground truth, its control experiment, and its deviation ledger from the row below.

Work typeGround truthControl experimentDeviation ledger
Integrate an external substrate, runtime, tracker, or servicethe vendor's own docs plus stock behaviorrun their vanilla quickstart on pinned versions with zero local code, before designingeach deviation from the documented flow, each justified; and every component you write that has a native counterpart in the substrate
Extend this projectthe repo's existing patterns and behavior specthe simplest version that satisfies acceptance, and why it is insufficienteach novelty introduced — new abstraction, dependency, or pattern
Greenfieldreference experience and domain prior arta walking skeletoneach deviation from the boring default, ~one novelty per change

The Extend row is already the repo's default discipline: behavior-first acceptance, RED -> GREEN, the smallest real change. The Integrate row is the one that is cheap to skip and expensive to have skipped — run the stock control experiment before you design, or you will re-plumb what the substrate already documents and inherit bugs you built yourself.

Trigger: the Integrate-row mechanics — the stock-quickstart control run and the deviation ledger from the documented flow — apply only to integration-class work (adopting or wiring in an external substrate, runtime, tracker, or service). Routine feature work on this project uses the Extend row and does not incur them.

A plan is done only when it passes the fresh-context test: a cold context, given the intent source alone, could execute it without the author's conversation. If execution needs facts that live only in the planning conversation, move them into the source before freezing.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.