Airlocks

Skill runsagents/airlocks

Verification gates for agent work, trust state, not words. Use when the user says "airlock", asks to verify a completion claim, or wants "merged/deployed/passing/published" proven rather than believed.From its SKILL.md

Install

npx -y skills add runsagents/airlocks

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

1.8 KB, 393 tokens by cl100k_base, as published. Nobody here has run it

Every completion claim from an agent, "merged", "deployed", "tests pass", "published", "fixed", is a claim, not a fact. Your job is to verify it against the system of record and report evidence.

The doctrine

Words are claims. State is evidence. Log lines, summaries, and "done" messages prove nothing.
Never grep for success words. Error messages contain the word "success" too. Parse exit codes and structured statuses only.
Ask the system of record, not the actor:
- Merged? The remote's default branch contains the commit: git merge-base --is-ancestor <sha> origin/<default>
- Deployed? The platform's API or the live endpoint reports the new version/build number.
- Tests pass? Run them on the merged commit, not the branch that claimed to pass.
- Published? The registry or store serves the new version to a clean client.
- Fixed? Reproduce the original failure case and watch it not happen.
Separate the verifier from the doer. Whoever performed the work does not get to verify it, use a fresh context or a different agent.
On mismatch: halt and report. Never auto-remediate unattended. File, don't fix, fixing without the human is how they wake up to confident nonsense.

Report format

For each claim, four lines:

CLAIM:            what the agent said happened
SYSTEM OF RECORD: what you queried
EVIDENCE:         the exact output/status you saw
VERDICT:          CONFIRMED / REFUTED / UNVERIFIABLE (and why)

By @runsagents, agents don't need supervision. They need airlocks.

What ships with it: 9 files

18.2 KB alongside SKILL.md, 2 of them executable

demo/

examples/

EXAMPLE.md1.1 KB

tests/

airlock-check.test.mjsruns3.3 KB

Gives 0 of the 12 instructions most quality gates skills give in 393 tokens

Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-07

Read the output and check the exit codein 54 of 1195, across 14 files
Verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
Identify the verification command proving the claimin 51 of 1195, across 12 files
Run the full verification commandin 50 of 1195, across 11 files
Verify output confirms the claimin 49 of 1195, across 12 files
Check version control diff after agent delegationin 46 of 1195, across 6 files
State claim with evidencein 44 of 1195, across 4 files
Run the test suitein 33 of 1195, across 26 files
Keep state in memory by defaultin 27 of 1195, across 6 files
Make prototype runnable with one commandin 26 of 1195, across 5 files
Produce a verification reportin 25 of 1195, across 14 files
Detect the package manager from lockfilesin 24 of 1195, across 5 files

Said here and by no other author read

verify completion claims against the system of record
parse exit codes and structured statuses only
query the system of record instead of the actor
run tests on the merged commit
halt and report on mismatch
file issues instead of fixing them unattended

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.