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 skills add backspace-shmackspace/claude-devkit --skill verification-before-completiongit clone --depth 1 https://github.com/backspace-shmackspace/claude-devkitWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/backspace-shmackspace/claude-devkit/verification-before-completion)<a href="https://agentmods.dev/skills/backspace-shmackspace/claude-devkit/verification-before-completion"><img src="https://agentmods.dev/badge/skills/backspace-shmackspace/claude-devkit/verification-before-completion/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/backspace-shmackspace/claude-devkit/verification-before-completion"><img src="https://agentmods.dev/badge/skills/backspace-shmackspace/claude-devkit/verification-before-completion.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 219 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00065 | $0.01910 |
| Opus 5.5 | $0.00026 | $0.00764 |
| Sonnet 5.5 | $0.00013 | $0.00382 |
| Haiku 4.5 | $0.00006 | $0.00191 |
Grade A, and why
verification-before-completion 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 1mo 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 — 220 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Verification Before Completion
Overview
Every completion claim requires fresh evidence. No exceptions.
The Iron Law
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE.
"I believe it works" is not evidence. "The tests passed at 14:32" is evidence.
If you have not run verification commands in this session and read their output, you have not verified.
The Gate Function
Before ANY claim that work is complete, execute this sequence:
1. IDENTIFY - What claims am I about to make?
2. RUN - Execute the verification commands
3. READ - Read the actual output (not what I expect to see)
4. VERIFY - Does the output support each claim?
5. CLAIM - Only now state completion, citing evidence
Every step is mandatory. Skipping any step means the gate has not been passed.
Step 1: IDENTIFY
List every claim you are about to make. Be specific:
NOT: "It works"
YES: "All 47 tests pass, the build succeeds, and the new endpoint returns 200"
Common implicit claims to surface:
- "Tests pass" -- which tests? All of them? Just the new ones?
- "Build succeeds" -- clean build or incremental?
- "Bug is fixed" -- does the specific reproduction case now succeed?
- "No regressions" -- have you run the full test suite, not just new tests?
- "Ready to merge" -- does it pass lint, type checking, and formatting too?
Step 2: RUN
Execute the actual verification commands. Not from memory. Not from a previous run.
REQUIRED: Run the commands NOW, in this session
REQUIRED: Use the Bash tool to execute them
REQUIRED: Run the FULL suite, not a subset (unless the user explicitly scoped it)
FORBIDDEN: Assuming tests still pass from earlier
FORBIDDEN: Referencing results from before your changes
FORBIDDEN: Skipping slow tests without stating you skipped them
Step 3: READ
Read the actual output. All of it.
REQUIRED: Read exit codes
REQUIRED: Read failure messages, not just success counts
REQUIRED: Read warnings (they may indicate real problems)
FORBIDDEN: Scanning for "X tests passed" and ignoring everything else
FORBIDDEN: Assuming green because the command "looked like it worked"
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.
- 1mo ago First seen · 220 lines · 65 tokens per session scan A 59769e6f8cde
verification-before-completion is a skill published in the GitHub repository backspace-shmackspace/claude-devkit (15 stars, last pushed 2mo ago), licensed MIT. It adds 65 tokens to every session and 1,910 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-08-30.
Other skills, from other repositories
architect
Produces Alon's design-doc system before code: SOURCEOFTRUTH.md, ARCHITECTUREROADMAP.md, TODOWORKFLOW.md, CLAUDE.md, plus a modular docs/architecture set for larger projects. Model first: data and invariants before framework, every invariant enforced at two boundaries, failure paths designed as deliberately as happy…
ponytail
Forces the laziest solution that actually works: YAGNI, reuse before new code, stdlib before custom, native platform before dependencies, one line before fifty. Three levels — lite (name the lazier alternative, build what's asked), full (ladder enforced, default), ultra (YAGNI extremist, challenge the requirement…
review-swarm
Local, free, multi-specialist review of a diff: parallel Claude subagents for correctness, security/trust boundaries, data/perf, architecture-altitude, ponytail-simplicity, and tests/failure-paths, each non-trivial finding adversarially verified before being kept, deduped, and ranked into a Blockers / Should-fix /…
ask-the-council
Convenes a panel of opinionated Claude advisors, each with a distinct mandate and a forbidden move so they genuinely diverge, then a Chairman synthesis that commits to one recommendation with the decisive tradeoff and the strongest dissent named.
prompt-generator
Turns a vague ask into a rigorous, grounded, token-efficient prompt for another agent or LLM: role and objective, testable done criteria, anti-hallucination and anti-tokenmaxing rules baked into the generated prompt itself, and strict agent discipline for coding-agent hand-offs.
up-to-date
Preflight sync and situational brief before repo work: fetches origin, reports ahead/behind divergence (with a no-upstream/detached-HEAD fallback), recent commits, open PRs and issues touching the work, and dirty-tree warnings, then names one recommended first action. Read-only by default — never pulls, rebases…