agentsclimarketplace

Automation shape routing

Skill boshu2/agentops/skills/automation-shape-routing

Front door for agent automation: choose inline, bounded fanout, reusable skill/workflow/gate, persistent agent-native workers, or explicit Gas City. Triggers: "build automation", "which orchestration shape", "should this use NTM".From its SKILL.md

Install
npx -y skills add boshu2/agentops --skill automation-shape-routing

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

One thing to look at

  • runs commandsInstructs the agent to run 1 command, including `printf '%s\n' "$verdict" | awk '...' (validator command for $verdict)`.

SKILL.md

5.8 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it

Automation Shape Routing

Choose the smallest execution shape that preserves the required evidence and control. This skill routes; it does not build or start a substrate.

Ordered routing works because each rung is strictly cheaper to operate than the next: if the smallest shape truly preserves the evidence and control the task needs, every larger shape can only add coordination cost, never correctness.

Named failure mode — substrate romance: routing to persistent workers because the topology is interesting, not because any deciding axis demands it.

Anti-pattern: starting the chosen substrate as part of routing "to save a step". Corrective: return the one-line verdict and let the owner start under its own authority.

Critical Constraints

  • Route only; do not start a substrate. Why: choosing an execution shape is a judgment step, while launching NTM, Agent Mail, or Gas City changes runtime state and requires separate operator authority.
  • Prefer the smallest shape that preserves evidence. Why: persistence and coordination add recovery and ownership costs that one-shot work cannot repay.
  • Partition write scopes before choosing concurrency. Why: a larger worker topology cannot make overlapping production writes safe.

Route in order

  1. One deliverable? Do it inline. Use a small in-session fresh-context fanout only when independent perspectives are the product. Do not create a reusable artifact for a one-off task.
  2. Reusable sequential procedure? Use skill-builder.
  3. Must-never-regress constraint? Route through operationalize to a gate.
  4. Fixed typed DAG, headless, no attach/steer? Use workflow-builder only where that runtime is explicitly selected and available.
  5. Persistent, attachable roles over caller-supplied packets? Use agent-native. NTM is the pane adapter; Agent Mail coordinates only explicitly selected live actors.
  6. Durable city of quests with GC-native supervision/store? Route to using-gc only when the operator explicitly selects Gas City. GC is not an automatic fallback or an ao runtime enum.

Deciding axes

AxisLightweight choiceEscalated choice
lifetimecurrent turnpersistent/attachable worker
topologyone writer or bounded fanoutdurable role graph
controlno mid-run steeringobserve/nudge/replace
outputone artifactreusable skill/workflow/gate
storecaller-owned packet setoperator-selected GC quest store
contentionone writerpartition, then Agent Mail reservation

Parallelism buys independence, not guaranteed speed. Refuse persistent orchestration for one-shot work, colliding write scopes, or a sequential chain that has no exploitable concurrency.

Handoff

Return exactly one of:

  • inline or bounded-fanout
  • skill-builder
  • workflow-builder
  • agent-native with a named reason persistent panes help
  • using-gc with explicit operator choice
  • operationalize:gate

Name the deciding axis and invoke the owner. Do not copy the delegated workflow into this router.

Output Specification

  • Artifact directory: stdout only; this routing decision creates no file.

  • Filename convention: none. Emit exactly one routing-verdict line.

  • Serialization/schema format: shape=<allowed-shape>; axis=<deciding-axis>; owner=<owning-skill> using one of the shapes listed under Handoff.

  • Owner mapping: inline and bounded-fanout use current-agent; skill-builder, workflow-builder, agent-native, and using-gc use the same value for owner; operationalize:gate uses operationalize.

  • Validator command: validate a captured $verdict as exactly one line with the declared shape/owner mapping:

    printf '%s\n' "$verdict" | awk '
      NR > 1 { extra = 1 }
      {
        valid = ($0 ~ /^shape=(inline|bounded-fanout); axis=[^;]+; owner=current-agent$/ ||
                 $0 ~ /^shape=skill-builder; axis=[^;]+; owner=skill-builder$/ ||
                 $0 ~ /^shape=workflow-builder; axis=[^;]+; owner=workflow-builder$/ ||
                 $0 ~ /^shape=agent-native; axis=[^;]+; owner=agent-native$/ ||
                 $0 ~ /^shape=using-gc; axis=[^;]+; owner=using-gc$/ ||
                 $0 ~ /^shape=operationalize:gate; axis=[^;]+; owner=operationalize$/)
      }
      END { exit !(NR == 1 && !extra && valid) }
    '
    
  • Downstream handoff: invoke the named owner only after returning the verdict; inline remains in the current agent and bounded-fanout remains in-session.

Quality Rubric

  • The verdict names exactly one allowed shape and one deciding axis.
  • Persistent or city-shaped routes cite the operator's explicit selection.
  • Concurrent routes state that production write scopes do not overlap.
  • The router delegates to the owner without copying or starting its workflow.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most automation workflows skills give in ~1.1k tokens

Counted across 813 of the 1,214 authors here whose files we hold, read 2026-09-06

  • Use conventional commit message formatin 38 of 813, across 34 files
  • Write tests before implementing codein 26 of 813, across 12 files
  • Achieve at least 80 percent test coveragein 24 of 813, across 11 files
  • Mock external dependencies for unit testsin 22 of 813, across 10 files
  • Read product marketing context before asking questionsin 21 of 813, across 6 files
  • Implement rollback plans for every deploymentin 21 of 813, across 9 files
  • Follow the arrange-act-assert patternin 20 of 813, across 9 files
  • Delete branches after mergingin 20 of 813, across 18 files
  • Test all edge cases and error scenariosin 20 of 813, across 8 files
  • Use semantic selectors for UI testsin 19 of 813, across 7 files
  • Define sequence type and audience contextin 18 of 813, across 5 files
  • Monitor feature drift and prediction distribution driftin 18 of 813, across 5 files

Said here and by no other author read

  • choose the smallest shape preserving required evidence
  • partition write scopes before choosing concurrency
  • return exactly one routing verdict line
  • name the deciding axis and owner
  • stop after returning the verdict
  • delegate to the owner without starting the workflow

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. 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.