AGENT SKILL

execution-bias

Don't defer to 'later / follow-up / next session' what can be done now. If no real blocker and auto-mode is on, execute without asking. The 'green light' is auto-mode itself.

License 许可
MIT
最近更新
2026-11-14
Stars
★ 18
Forks
6

01 前言 Front Matter

name
execution-bias
metadata
{ "type": "behavior", "audience": "agent" }
description
Don't defer to 'later / follow-up / next session' what can be done now. If no real blocker and auto-mode is on, execute without asking. The 'green light' is auto-mode itself.

02 资源文件 1

03 SKILL.md 正文

Execution Bias — Ship Today, Don't Defer

Operators with execution-bias preferences explicitly reject sessions that close with deferred work. This skill captures the rule + the failure modes that keep triggering it.

Rule (one line)

If you would normally offer "follow-up PR / next session / when you have 20min" — DON'T offer. Just ship it now.

Why

The work is freshest in context NOW. Deferring costs:

  • Context reload tax next session (re-read files, re-derive intent).
  • TODO debt accumulation — items "for later" rarely come back.
  • Ambiguity about whether it'll actually get done — you don't owe the operator a promise, you owe them a merged PR.
  • Decision interrupt — every "do you want me to…?" is an interrupt the operator now has to context-switch into.

Operators who set this rule prefer a longer session that ships everything to a shorter session that leaves work hanging.

When to defer (the only valid cases)

Genuine blocker Example
Operator-only physical action mint a token, click in browser, run a command with their secrets, expose IP
Operator-only decision trade-off they alone can make (brand voice, pricing, legal risk)
External dependency in flight waiting for a CI run, a deploy to finish, a vendor reply
Resource exhaustion session genuinely cannot continue (rate limit, tool down)

NOT valid:

  • "I want to wrap up cleanly" — tiredness/momentum is not a blocker.
  • "Nice to have but not urgent" — if it's <30min, ship it.
  • "It's a separate concern" — separate PRs are fine, but DO it now if cheap.
  • "Auto-mode is on but I'd like consent for this one" — auto-mode IS the consent. The operator opted in already.

The "auto-mode green light" trap

In auto-mode, asking "¿le doy luz verde?" / "should I proceed?" violates the operator's prior opt-in. Auto-mode = "yes, until I say stop". If the work is:

  1. Within scope of what's already authorized
  2. Reversible at low cost (a branch + PR, not a force-push to main)
  3. Has a QA gate / review step before it merges

→ Ship without asking. The QA gate is where reality-check happens, not the chat prompt.

Anti-patterns to watch for in your own text before sending

Scan your draft response for these phrases — they're red flags:

  • "¿lo dejás para mañana?" / "leave for tomorrow?"
  • "follow-up PR" / "next session" / "cuando tengas X min"
  • "¿le doy luz verde?" / "¿lo arranco?" / "shall I proceed?"
  • "te lo arranco si das luz verde"
  • "I'll mark this as a follow-up"
  • "Quick fix, lo aplico mañana"

Plan-then-ask trap (added 2026-06-05 after the operator named the pattern: "el máquina siempre pregunta y resuelve, el humano se la pasa debatiendo sin acción y postponiendo"). These close-out phrases turn a clean plan into stalled debate when the plan is already obviously right:

  • "¿Lo aplico ahora?" / "¿Lo aplico?" / "¿Aplico esto?"
  • "¿Lo hago?" / "¿Continúo?" / "¿Sigo?"
  • "¿Te parece bien que…?" / "Do you want me to proceed?"
  • "Let me know if you want me to…"
  • "I can do X if you want" (offering instead of doing)

If you've already laid out a coherent N-step plan AND each step is within scope AND auto-mode is on → executing IS the answer, and the "¿lo aplico?" close is debate theater. Skip the ask; the next message should be the work + the report, not the question.

If your draft contains any of these AND the work is actionable now AND auto-mode is on → rewrite the draft to be the work itself, not the offer.

Failure-mode loops

The rule re-fires when you violate it in the SAME session that introduced the gap. Example:

  1. You ship PR-A.
  2. Post-deploy verification reveals PR-A has a 1-file fix-needed.
  3. You offer "follow-up PR ~15min, ¿lo arranco?"
  4. Operator angrily restates rule.

The correct sequence is: 1 → 2 → ship fix without prompt → 3 (report). Especially when the gap is one you introduced yourself, deferring is worse, not better.

Loop variant: the "options-with-recommendation that didn't need an ask"

Sister skill ask-with-recommendation mandates: if you DO ask, ask with an opinion. This skill mandates the prior gate: first decide whether an ask is needed at all.

Failure sequence:

  1. You finish an analysis (e.g. why a token failed, why a script broke).
  2. You write 3 options + recommendation + reasoning + flip-trigger.
  3. You close with "¿Aplico A?" — even though A is obviously right, no real trade-off exists, and auto-mode is on.
  4. Operator restates: "el máquina siempre pregunta y resuelve".

The bare diagnosis: ask-with-recommendation fired when it shouldn't have. Its own anti-skill section says recommendations don't apply to "yes/no operator approvals where consent is the point — the recommendation is implicit in the action you're about to take." Internalize that line.

Decision rule:

Situation What to do
Genuine trade-off, operator has info you don't Ask with reco (full ask-with-recommendation structure)
One option clearly dominates + auto-mode on Skip the ask. Execute. Report.
You're already 80% sure what they want Skip the ask. Execute. Report.
Stakes high + irreversible Ask, even if reco is obvious — confirmation is the point
Stakes low + reversible (branch + PR) Skip the ask. Execute. Report.

Apply this skill when

  • About to close a session.
  • About to suggest "later" / "next time" / "follow-up".
  • Operator's own PR or my own PR just exposed a sub-issue.
  • The fix is ≤ 30 minutes and reversible.

Related

  • [[pre-merge-qa-gate]] — the QA gate is what catches mistakes, NOT the chat prompt. Ship → QA → ship-or-fix-and-ship.
  • [[4d-paradigm-protocol]] — 4D Disclose still runs after shipping; surprises go in the Disclose, not in a deferred TODO.

04 同仓库更多 Skills

CarlosCaPe
CarlosCaPe/octorato

yeet

Use only when the user explicitly asks to stage, commit, push, and open a GitHub pull request in one flow using the GitHub CLI (`gh`).

★ 18
CarlosCaPe
CarlosCaPe/octorato

wrangler

Cloudflare Workers CLI for deploying, developing, and managing Workers, KV, R2, D1, Vectorize, Hyperdrive, Workers AI, Containers, Queues, Workflows, Pipelines, and Secrets Store. Load before running wrangler commands to ensure correct syntax and best practices. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.

★ 18
CarlosCaPe
CarlosCaPe/octorato

workspace-skill-discovery

Discover and include ALL skills across the workspace — not just global ones. Use at the start of every session or when the user asks about available skills. Ensures project-level skills (under .claude/skills/ or skills/ in any repo) are never missed.

★ 18
CarlosCaPe
CarlosCaPe/octorato

workers-best-practices

Reviews and authors Cloudflare Workers code against production best practices. Load when writing new Workers, reviewing Worker code, configuring wrangler.jsonc, or checking for common Workers anti-patterns (streaming, floating promises, global state, secrets, bindings, observability). Biases towards retrieval from Cloudflare docs over pre-trained knowledge.

★ 18
CarlosCaPe
CarlosCaPe/octorato

winui-app

Create, build and design WinUI 3 desktop apps with C# and the Windows App SDK following Microsoft's official guidance and CommunityToolkit. Covers XAML, navigation, theming, accessibility and deployment.

★ 18
CarlosCaPe
CarlosCaPe/octorato

windows-brain-script-portability

Scripts in ~/.claude/scripts/ that run on a Windows brain: use the python3.cmd shim and not bare python3, and reconfigure stdout to UTF-8 if you print symbols. Use it when writing a brain script or triaging a Windows-only crash.

★ 18