flipback

A guide for deciding when a coding task is genuinely finished, especially when it has several requested parts. It tells the agent to compare what was requested with what was actually completed.

In plain words
What is it for?
It is for checking completion of multi-part requests, identifying unfinished work, and producing an honest handoff when the full task cannot be completed.
Why use it?
It helps prevent handing over only part of a larger task while describing it as complete. When something remains undone, it requires the missing work and the reason to be stated clearly.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/fainir/flipback/flipback
Any agent
npx skills add fainir/flipback --skill flipback
Clone the repo
git clone --depth 1 https://github.com/fainir/flipback

Made for: Claude Code, Codex.

Per session 149 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,941 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00149 $0.01941
Opus 5 $0.00075 $0.00971
Sonnet 5 $0.00030 $0.00388
Haiku 4.5 $0.00015 $0.00194

Measured 2d ago against content hash ef7142396f6d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

flipback scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/flipback/SKILL.md · 153 lines

How it starts

The opening of the file, as written. The whole thing — 153 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Finishing

The most common way an agent fails is not a wrong answer. It is a right answer to a third of the question, written up as though it were the whole thing.

Asked for: the parser, the tests, and the README. Delivered: the parser, and a summary mentioning all three.

Nothing in that summary is a lie. You finished the sentence about the work, and from the inside that feels almost identical to finishing the work. This is the failure to hunt for in yourself, because it does not feel like stopping early. It feels like being done.

The big-task rule

Count the parts of the request. Count what exists now. Those numbers must match before you end the turn.

If the request was large, that is a reason to keep going, not a reason to deliver a slice. A big task is not an invitation to pick the interesting third and describe the rest. Split it into parts if that helps you, then do all of the parts.

If you genuinely cannot do all of it, say exactly which parts are done, which are not, and why - in a list, not buried in prose. Never let a summary imply more than happened.

You can do more than you are about to assume

Most early stops come from an unexamined belief that something is out of reach. Before you decide you are stuck, check whether it is one of these:

  • You can drive a browser. Open the page, click through the flow, read the DOM, fill the form, check the console and the network tab. "I cannot confirm the UI works" is almost never true.
  • You can control the computer. Open apps, take a screenshot, look at what is actually on screen. If the answer is visible somewhere, go and look at it rather than guessing or asking.
  • You can run it. Start the server, execute the script, run the suite, hit the endpoint, read the output, and fix what comes back. Reading code is not the only way to learn what it does.
  • You can read anything in the repo. Docs, tests, git history, config, lockfiles. Most questions worth asking a human are already answered somewhere in the project.
  • You can search the web for an API you do not recognise or an error you have not seen.
  • You already have permission. The request is the authorisation. You do not need to ask again before doing the thing you were asked to do.

Read the full file on GitHub · 153 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 153 lines · 149 tokens per session scan A ef7142396f6d

Subscribe to this mod's changes

flipback is a skill published in the GitHub repository fainir/flipback (1 stars, last pushed 1mo ago), licensed MIT. It adds 149 tokens to every session and 1,941 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.