agentsclimarketplace

Dos plan price

Skill anthony-chaudhary/fak/.claude/skills/dos-plan-price

Price a proposed multi-agent fan-out before launching workers. Use when a packet, goal fleet, or hand-written plan would run several agents over declared file trees and you need DOS to catch collisions before any worker starts.From its SKILL.md

Install
npx -y skills add anthony-chaudhary/fak --skill dos-plan-price

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

SKILL.md

3.3 KB, as published. Nobody here has run it

dos-plan-price - price fan-out before launch

Use this before launching more than one worker from a dos-next-up packet, a goal fleet, or an operator-supplied partition. dos arbitrate is still the floor at worker acquire time; this skill is the earlier pricing pass that avoids spending launches on a partition whose trees already collide.

Load-Bearing Rule

Do not launch a colliding fan-out. Price the proposed worker trees first, then either launch the safe set or narrow the partition and re-price.

Step 0 - Discover Workspace Geometry

dos doctor --workspace . --json

Read lanes.trees and resolve lane names to their configured trees. Do not hardcode lane trees. A worker whose tree is unknown must stay unknown in the partition; unknown scope is a collision risk, not permission to launch.

Step 1 - Build The Proposed Partition

Create one entry per worker:

{"name": "worker-name", "tree": ["repo-relative/glob/**"]}

Sources can be a dos-next-up dispatch list, a dos-goal-fleet wave, or an operator's explicit worker list. If a worker has no named tree, record an empty tree and treat the result as unsafe until the scope is made explicit.

Step 2 - Price The Partition

There is no first-party dos price-plan CLI verb yet. Use the shipped interim example if available, and log the gap instead of pretending the verb exists:

python examples/plan_price/plan_price.py --json

For a custom partition, import the example API from examples/plan_price if it exists in the current workspace or in the installed DOS kernel checkout:

from plan_price import Agent, price_plan

price = price_plan([Agent(name, tree) for name, tree in proposed])

Log this every run while the first-party verb is absent:

log: predictive fan-out pricing is example-backed pending a first-party dos price-plan verb; dos arbitrate remains the acquire-time floor.

Step 3 - Act On The Price

  • collisions == 0: launch the full fan-out, then have every worker call dos arbitrate at acquire time.
  • collisions > 0: do not launch the colliding plan. Launch only safe_now, or narrow the colliding trees and re-run the price.
  • unknown tree: stop and name the worker's scope before launch, unless the operator explicitly chooses a serial run.

Return the action in a machine-readable final line:

Action: LAUNCH_ALL | LAUNCH_SAFE_SET | REPARTITION_AND_REPRICE | SERIALIZE_UNKNOWN_SCOPE

Step 4 - Keep The Reactive Floor

Pricing is a forward estimate over declared scope. It does not hold a lease and does not replace dos arbitrate:

dos arbitrate --workspace . --lane <lane> --kind cluster

Honor the acquire/refuse verdict. Do not use pricing as a reason to force an arbiter refusal.

Anti-Patterns

  • Launching a fan-out after the price reports a collision.
  • Treating the price as a lease.
  • Skipping dos arbitrate because the price was clean.
  • Resolving unknown scope to **/* silently and launching anyway.
  • Hardcoding lane names or trees instead of reading dos doctor --json.

Keep looking

Skills are one crate of 615,528. 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.