# Max Effort

> Posture skill activated by `/max-effort` (single-task) or `/max-effort sustained` (session). Adversarial review of the work in front of you — does it answer the real goal, and would it survive someone trying to break it? Sustained mode persists until an explicit off-signal from the user.

- **Type:** Skill
- **Install:** `agentstack add skill-ikkeseb-skills-max-effort`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [ikkeseb](https://agentstack.voostack.com/s/ikkeseb)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [ikkeseb](https://github.com/ikkeseb)
- **Source:** https://github.com/ikkeseb/skills/tree/main/skills/max-effort

## Install

```sh
agentstack add skill-ikkeseb-skills-max-effort
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# max-effort

Posture for high-stakes / irreversible work. The load-bearing move is *adversarial*: with the work in front of you, ask whether it answers the real goal and would survive someone trying to break it — then try to break it before you ship.

Not a "be careful" prompt, and not an engine for producing more work. This skill *tears down* — it assumes the work might be wrong and tries to prove it.

## Modes

- `/max-effort ` — single-task, one review round.
- `/max-effort sustained` — session posture, persists until explicit off-signal.

## Off-signal (sustained)

Only an explicit user signal drops it: `max-effort off`, `stop max-effort`, or any unambiguous stop/return signal in any language. Mid-sustained questions or redirects don't drop it. New session starts fresh.

## Preamble — first response after `/max-effort`

Open with the marker: `[max-effort]` (single) or `[max-effort sustained]`. Surface task ambiguity in one batch only if it's genuinely unclear *what* is asked — no "should I be thorough?" padding. Then go.

## The pass

Always run, in this turn, against the work however it was produced:

- **Goal-match.** Does it answer the question asked, or a neighbor it drifted into?
- **Break it.** Red-team what's irreversible or externally-visible — spawn skeptics to tear down when the stakes warrant it. Confidence isn't evidence, yours or a subagent's.
- **Check load-bearing claims against source-of-truth.** Read the cited file, open the URL, run the tests/build/repro, look at the browser — task-specific, the CLAUDE.md verification bar not relaxed.
- **Resolve conflicts.** Don't paper over disagreement with "both have a point" — navigate it.
- **Calibrate.** Surface a gap rather than claim coverage you didn't deliver. If you haven't checked one source-of-truth artifact this turn, the pass didn't happen.

Trivial task with nothing load-bearing to check: say so once in the preamble, ask whether to proceed.

## Stacking

- **`full-send`**: full-send owns generation; max-effort owns the adversarial pass on the converged result. Build wide, then tear down — in that order.
- **`afk`**: the preamble does not run (asking is forbidden under AFK) — default to single-task, log it, and run the pass unchanged within AFK's blast-radius rules.

## Final summary

Inline, marker-led — no persistent log (cross-session "thorough on X" state breeds false certainty). Not a recap of the work above: just the marker and what's still open.

In sustained mode the marker is load-bearing — it's how the posture survives context compression, so it appears on every turn that emits a summary. After it, **open items only**: one `🚩` line per unresolved conflict, unverified claim, assumption leaned on, or decision waiting on you. Always at least one `🚩`, or `✅ nothing open` when there genuinely is nothing — the anti-cascade anchor.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [ikkeseb](https://github.com/ikkeseb)
- **Source:** [ikkeseb/skills](https://github.com/ikkeseb/skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-ikkeseb-skills-max-effort
- Seller: https://agentstack.voostack.com/s/ikkeseb
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.