Levyra mode
SkillDev toolsUse only for explicit owner execution cues such as VAI, PROCEDI, INTERVIENI, FAI TU, FALLO TU, or requests to open/create a PR. Keep execution fast and low-noise while preserving full Levyra engineering rigor.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Levyra mode skill
What this skill tells your AI
The instructions your AI receives, as published by luc4n3x/levyra-deepsound in .agents/skills/levyra-mode/SKILL.md and read by ahel’s review.
Purpose
Make owner-directed work faster to execute and easier to follow. Save tokens by
removing redundant context and narration, never by reducing technical analysis.
Root/scoped AGENTS.md, always-on guards, domain skills, validation, review, and
publication controls remain authoritative.
Full engineering rigor
For every code, build, workflow, schema, or agent-config change:
- inspect the current owner/control flow and nearby tests before editing;
- identify the root cause or exact implementation contract instead of stacking guesses;
- make the smallest coherent change and preserve lifecycle, cancellation, concurrency, data, security, and compatibility semantics that apply;
- run focused validation after the latest material edit and keep failures classified from evidence;
- inspect the complete final diff, run
git diff --check, and perform the required code-review gate before delivery or publication.
Never skip source inspection, a necessary test, decisive diagnostics, security review, or final code review to save tokens. Token efficiency is allowed to remove repetition, not engineering depth.
Token discipline
- Load only skills that materially affect the current task.
- Search symbols/paths first and read bounded source ranges before whole files.
- Do not reread unchanged instructions, files, logs, or evidence already in context.
- Compact routine successful command output; preserve or rerun raw when exact evidence can change the diagnosis.
- Keep updates to new evidence, current action, and real blockers.
- Carry forward compact verified handoffs, not exploratory chatter or disproved hypotheses.
Execution shape
- Start with the action, result, or blocker; avoid generic preamble.
- Keep visible multi-step plans to at most five concrete steps.
- Perform authorized tool/edit/review work directly instead of handing it back to the owner.
- Finish the requested scope before unrelated cleanup.
- Report exact delivery and validation state; never manufacture completion or timing.
- Commit, push, PR, merge, release, deployment, version changes, and repository settings still require the authorization defined by Levyra.
Provenance
The action-first interaction ideas are selectively adapted from the MIT-licensed
ayghri/i-have-adhd project. Levyra keeps only the execution ergonomics and does
not add diagnosis-specific behavior or an external runtime dependency.
Signals
- GitHub stars
- 164
- Forks
- 1
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
levyra-mode- Source
- github.com/luc4n3x/levyra-deepsound