Install
$ agentstack add skill-gulmezeren2-byte-andon-verify-with-andon ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Verify analysis with andon
You have just produced (or edited) an analysis artifact: a report workbook, a summary table, a set of claimed figures. Before presenting it, verify it with andon. The verification is deterministic arithmetic — your own confidence in the numbers is not a substitute for it.
The discipline
- Write the spec while the derivation is fresh. For every number you put in the
artifact, you know which source rows and filters produced it. Encode exactly that as a check. If you cannot write the check, you do not actually know where the number came from — treat that as a defect in the analysis, not in the spec.
- Cover four layers, in this order:
schema.*— the assumptions about the source data (unique keys, allowed values,
date continuity);
reconcile.*— every headline figure against the source, with the same filters
you used to derive it;
internal.*— totals vs parts, percentages vs 100, derived columns vs their
formulas;
excel.integrity— if the artifact is a workbook.
- Run and read:
`` andon run andon.yaml --json ``
Exit 0 = all pass. Exit 1 = at least one FAIL: the artifact is wrong or the spec is — find out which before touching anything else. Exit 2 = REVIEW flags: report them to the user verbatim; do not resolve them yourself. Exit 3 = nothing was verified (a check could not run, or every check was skipped): fix the spec or say plainly that the claim is unverified. Exit 4 = broken spec.
- Report the verdict with the artifact. Include the counts (passed/failed/review)
and the honesty block's "Never read" list. A clean verdict is part of the deliverable; a dirty one is the deliverable.
Hard rules
- Never loosen a tolerance, add a
skip:, or narrow awhere:to make a check pass
without stating in your response exactly what you changed and why. A silently weakened spec is worse than no spec.
- Never present a number the spec does not cover as if it were verified. Say which
figures are covered and which are not.
- If a cell read fails with "no cached value" (library-written workbook), point the
check at the underlying data range instead of the formula cell.
- Prefer failing loudly over passing quietly: when unsure between two ways to write a
check, write the stricter one and let the human relax it.
Spec syntax
Run andon init for a commented starter, or see the full reference: https://github.com/gulmezeren2-byte/andon/blob/main/docs/checks.md
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: gulmezeren2-byte
- Source: gulmezeren2-byte/andon
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.