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.
npx agentmods add skills/everywan-dev/claude-code-engineering/verify-before-saying-donenpx skills add everywan-dev/claude-code-engineering --skill verify-before-saying-donegit clone --depth 1 https://github.com/everywan-dev/claude-code-engineeringWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00034 | $0.01406 |
| Opus 5 | $0.00017 | $0.00703 |
| Sonnet 5 | $0.00007 | $0.00281 |
| Haiku 4.5 | $0.00003 | $0.00141 |
Grade A, and why
verify-before-saying-done 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.
How it starts
The opening of the file, as written. The whole thing — 131 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Verify before saying done
A check that cannot fail is not a check.
Before the words "done", "fixed", "working" or "deployed" leave you, you have to be able to answer one question:
What did I run that, if this were broken, would have returned something different?
If there is no answer, you have a reading, not a verification. Say so instead.
When to use this
- You are about to close a task, comment on a ticket, or answer "is it up?"
- You changed a config, restarted a service, edited a file inside a container
- You fixed something visual
- Someone asks you to confirm a state you have not observed since you changed it
The procedure
1. Write the claim as one sentence. Not "the deploy went fine" — "the API returns the new schema version on all four nodes". A vague claim cannot be falsified, which is why vague claims are the ones that survive.
2. Name the failure mode you are ruling out. "If this were broken, what exactly would be wrong?" The answer picks your instrument. Ruling out "the service did not pick up the config" needs the process's own view of its config, not the file on disk.
3. Pick an instrument that observes the claim, not a neighbour of it. Most false verifications are a measurement of something adjacent:
| You measured | You claimed | The gap |
|---|---|---|
| The file parses | The service loaded it | Loading can be rejected after parsing |
| The file on disk changed | The process reads the change | Different inode, different file |
| The port answers | The application works | Anything can answer a port |
| It works in staging | It works in production | Different components, different risk |
The CSS says red |
The user sees red | Something else wins the cascade |
4. Run it and keep the raw output. Not your summary of it. The count, the status line, the response body, the screenshot. Summaries are where the optimism gets in.
5. Prove the check is live. Break something on purpose — point it at a hostname that does not exist, at a node that is down — and confirm it goes red. A green result from a check you never saw go red tells you nothing about the system, only about the check.
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.
- 2d ago First seen · 131 lines · 34 tokens per session scan A 2fcf89c7e8ef
verify-before-saying-done is a skill published in the GitHub repository everywan-dev/claude-code-engineering (2 stars, last pushed 14d ago), licensed Apache-2.0. It adds 34 tokens to every session and 1,406 once invoked, about $0.0002 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.
Other skills, from other repositories
claude-project-manifest
Creates annotated bibliography-style manifests for Claude projects, tracking files, conversation threads, and relationships with unique IDs and annotations. Use when documenting project contents, creating file inventories, tracking conversation history, or building navigable knowledge maps.
proofpress
Preserve or inspect decision history for Markdown and static HTML knowledge artifacts with Proofpress. Use when creating or accepting a meaningful revision to an ADR, design doc, PRD, README, strategy note, portable .md artifact, or static .html/.htm artifact in a configured repository, or when receiving a file…
llm-wiki
Maintain a personal team knowledge base using the LLM Wiki pattern — incremental ingest, query, and lint operations on a layered wiki architecture.
PBE Extractor
Extract invariant principles from any text — find the ideas that survive rephrasing.
Principle Synthesizer
Synthesize invariant principles from 3+ sources — find the core that survives across all expressions.
wiki-lint
Run a deterministic, read-only health check on an Obsidian wiki. Use for lint, vault health check, audit wiki health, find orphans, find dead links, frontmatter audit, provenance audit, or wiki audit. Reports graph, link, frontmatter, provenance-ledger, empty-section, and stale-index findings; it does not reason…