AGENT SKILL

setup-pstack

Configure which models pstack uses per role and at what reasoning budget. Detects your available models and writes the shared model registry that overrides the skill defaults. Use for /setup-pstack, "configure pstack models", "pstack budget", or changing pstack's model choices.

License 许可
MIT
最近更新
2026-10-30
Stars
★ 0
Forks
0

01 前言 Front Matter

name
setup-pstack
description
Configure which models pstack uses per role and at what reasoning budget. Detects your available models and writes the shared model registry that overrides the skill defaults. Use for /setup-pstack, "configure pstack models", "pstack budget", or changing pstack's model choices.
disable-model-invocation
true

03 SKILL.md 正文

Setup pstack

Write ~/.config/agent-workflows/models.toml, the model registry that sets pstack's model per role. Claude Code, Codex, and Grok all read it, through aw-run.

Steps

1. Detect available models

Detect only for the CLIs on PATH. Codex: run codex debug models and take the slug of each entry with "visibility": "list", with its supported_reasoning_levels. Claude Code: the CLI accepts the aliases opus, sonnet, haiku, and fable, or a full model name. It has no listing command, so treat those aliases as detected and ask the user to paste any full name they want. Grok: run grok models and take each model id it lists. Never write a model you have not confirmed is available. The value inherit is always valid even though it is not a detected model.

2. Load current state

The default role-to-model mapping is the registry shape shown in step 5 below. If ~/.config/agent-workflows/models.toml already exists, read it and treat its budget and its roles as the current choices. Otherwise start from those defaults. A role not in step 5, such as how_critics, is from a retired role. Drop it.

3. Budget, map, and confirm

(a) Ask for a budget. Ask with the CLI's structured question tool when it has one, otherwise in plain text. Offer these four options with these exact labels, and name the current budget when the registry records one. With no registry, say that large matches the skill defaults.

  • unlimited — max reasoning
  • large — xhigh reasoning
  • medium — high reasoning
  • small — medium reasoning

(b) Apply it. The registry's budget sets the starting effort of every role, panel entries included, to max, xhigh, high, or medium, on the ladder max > xhigh > high > medium > low. All three CLIs accept every level on that ladder. An entry may set its own effort and speed as model:effort or model:effort:fast, which overrides the budget. Only Codex models have a fast tier. aw-run applies both. inherit does not change.

(c) Show the roles and confirm. Show every role with its model, provider, and effort at the chosen budget, marking any model not in the detected set as needing a choice. Also list each role step 2 dropped. Ask whether to accept as-is or change specific roles, offering the detected models plus inherit (the role runs on the parent session's model) as the options. For panel roles (arena_runners, architect_runners, interrogate_reviewers) the value is a list, and one subagent runs per entry, inherit entries included, so the list length sets the count. arena_cross_judge_pool is also a list, but Arena selects one value from it whose provider differs from the parent's when possible. swarm_workers is the default model for every worker unless a race or comparison assigns another model per arm.

Each model entry names its provider, claude, codex, or grok, and that CLI's subscription pays for every call on it. Show how many roles each provider carries so the user can balance the subscriptions. Each model also names a fallback list of models on other providers. aw-run resolves a model whose CLI is not installed to the first installed model in that list, and retries down the list when a provider reports a usage limit.

4. Validate

Every model written must be in the detected set. inherit always passes. Each fallback must name another model in the registry. If a chosen model is not available, stop and ask again. Then run aw-run --resolve <role> for every role and fix any error it prints.

5. Write the registry

Write ~/.config/agent-workflows/models.toml with a budget line, one [models.<name>] table per model, and one line per role, using the role names the workflow skill uses. Overwrite the whole file so re-runs stay idempotent. Shape:

# pstack model configuration. One line per role. Delete a role to fall back to the skill default.
# "inherit" as a value: the role runs on the parent session's model. Inherit entries in a panel list still count toward its fan-out.
budget = "large"

[models.opus]
provider = "claude"
model = "opus"
fallback = ["sol", "grok"]

[models.sol]
provider = "codex"
model = "gpt-6.1-sol"
fallback = ["grok", "opus"]

[models.grok]
provider = "grok"
model = "grok-4.7"
fallback = ["sol", "opus"]

[roles]
feature_refactoring = "sol"
bug_fix = "sol"
perf_issue = "sol"
hillclimb = "sol"
judgment_prose = "opus"
hardest_tasks = "opus"
how_explorer = "sol"
how_explainer = "opus"
why_investigators = "sol"
why_synthesizer = "opus"
reflect_tooling = "sol"
reflect_judgment = "opus"
arena_runners = ["opus", "sol"]
arena_cross_judge_pool = ["opus", "sol"]
swarm_workers = "sol"
architect_runners = ["opus", "sol"]
interrogate_reviewers = ["opus", "sol"]

6. Confirm

Tell the user the registry was written and that it applies to the next spawn in every CLI. Re-running this skill updates it.

7. Offer a verification skill (optional)

Check whether the project has a way to drive the real app for proof (a verify-* skill, or an existing harness). If not, offer once: "want a project-local verification skill, so agents can drive the app the way a user does and prove changes work? I can generate one with /create-verification-skill." On yes, invoke /create-verification-skill (resolves wherever pstack is installed: project or user skills). On no, move on without pushing.

04 同仓库更多 Skills

shreshthkapai
shreshthkapai/agent-workflows

workflow

Concise, detailed responses, deliberate subagents, unslopped prose, simple code, and verified work, with each task routed to a playbook. Use for /workflow, or for requests to work in this style. `aw skills always workflow` loads it into every session.

★ 0
shreshthkapai
shreshthkapai/agent-workflows

why

Use for 'why does X work this way', 'why we picked Y', design rationale, regressions, postmortems, or data-backed thresholds. Discovers available MCPs and queries each evidence category (source control, issue tracker, long-form docs, real-time chat, infrastructure observability, error tracking, product analytics warehouse) in parallel, then returns a cited read on decisions and tradeoffs. Use how for runtime behavior.

★ 0
shreshthkapai
shreshthkapai/agent-workflows

unslop

Cut AI tells from any writing.

★ 0
shreshthkapai
shreshthkapai/agent-workflows

typescript-best-practices

TypeScript best practices. Use when reading or editing any .ts or .tsx file.

★ 0
shreshthkapai
shreshthkapai/agent-workflows

technical-writing

Layered technical-writing standard: Diátaxis structure, Google developer style sentences, STE instruction rules, Global English syntax. Use for /technical-writing or when writing or reviewing docs, RFCs, readmes, PR descriptions, or commit messages.

★ 0
shreshthkapai
shreshthkapai/agent-workflows

teach

Explain a body of work plainly so a person actually understands it. Runs the `how` and `why` skills and weaves what they find into one clear explanation. Use for 'teach me this', 'help me really understand X', 'explain this change or subsystem to me'.

★ 0