- Home
- Skills
- Testing & Debugging
- Luciazero
More screenshots
Works with the AI Agents you already use
Luciazero
Make coding agents prove their work: a verify-before-done operating mode with closeout, debug, plan, setup, retro, experiment and review procedures.
Free
Luciazero
Example session with this skill installed
Use Luciazero's /debug: ./test.sh exits 0 on macOS even when a file in tests/gates/ has a syntax error. Find the cause, fix it, and prove the fix.
- Read your context and instructions
- Compiled the luciazero
From a real Luciazero session (PR #18 in the Luciazero repo). The prompt is paraphrased; every output below is quoted from the runs, and the before/after table was re-run on 2026-10-14.
Summary
Bash 3.2, the macOS /bin/bash, returns 0 from source when the sourced file has a syntax error, so a broken gate let ./test.sh finish green. In --full, a broken full-only gate even printed PASS all checks green. The dispatcher now parses each gate before sourcing it, and a regression test proves every tier fails on a malformed gate.
Hypothesis ledger
H1: `source` of a file with a syntax error exits 0 under Bash 3.2
refutable by: run test.sh over a stub gate containing `if then` with /bin/bash
→ confirmed (rc=0 in every tier)
Reproduction — /bin/bash 3.2.57, stub gate if then
| Tier | Before the fix | After the fix |
|---|---|---|
--discipline | rc=0, no failure reported | rc=2, FAIL: gate syntax syntax |
--fast | rc=0, no failure reported | rc=2, FAIL: gate syntax syntax |
--full | rc=0, no failure reported | rc=2, FAIL: gate syntax syntax |
Fix — test.sh
# Bash 3.2 can exit zero on a syntax error in a sourced file. Parse first,
# with the shell that will source it.
"${BASH}" -n "${ROOT}/tests/gates/${GATE_NAME}.sh" || {
local GATE_RC=$?
echo "FAIL: gate ${GATE_NAME} syntax" >&2
return "${GATE_RC}"
}
Regression test: red, then green
- Test added first, fix not yet applied:
./test.sh --full→ exit 1,FAIL: --discipline accepted a syntax error in a sourced gate - With the fix: exit 0,
ok malformed sourced gates return non-zero in every tier, a full-only gate included→PASS all checks green
Done report
Done: test.sh fails when a gate does not parse, in every tier
Proof: ./test.sh --full → PASS all checks green
Not covered: gates that parse but misbehave at run time; their own exit status still decides
Left out: nothing
Next steps
- Merged as PR #18 after CI and an independent review passed.
- Keep the pre-source parse check; the tiers gate fails if it is removed.
Connects securely to your tools. The creator never sees your data.
About this skill
Luciazero makes coding agents prove their work instead of reporting it. The operating mode requires a command with an exit code before any "done", treats a missing verify command as the first bug, never lets a check be weakened to reach green, and makes the agent stop and ask before destructive or outward-facing actions.
Included
- The Luciazero operating mode (9 rules)
- Procedures:
/donecloseout,/ready(verify-command setup),/debug(hypothesis ledger),/plan,/retro,/experiment,/show - An adversarial reviewer prompt that tries to refute a diff rather than approve it
Not included
Helper scripts, hooks, the Agent Bus, /lucia-relay and /bisect are not in this package; they ship with the full open-source Luciazero pack (MIT): github.com/ohm41321/luciazero. This package says how to do each scripted step by hand.
How to use
Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scanned
Verified clean 6 days ago
- Free to download with an account
Requires