Skills

One hundred twelve skills is far more than any one task needs — and that's the point. Nobody reads them all. Each skill does exactly one job and carries a why, a measured token badge, and verify checks. They are written for our philosophy — code for agents to read, interfaces for humans to judge — not for human maintainers. Every skill is backed by 15+ years of production experience, industry professionals, and published studies — from OWASP and NIST to the SRE book and peer-reviewed research.

Progressive disclosure

How it works: 1. the bank map routes your task to an area; 2. your agent fetches only the matched skills with its key; 3. install only what matches and work from local copies. Small skills plus one map keep every session’s context small by design — the whole bank is about 21.0k tokens against typical task packs of 2.5–3.6k. It’s the same pattern Anthropic describes as progressive disclosure: agents pulling in context as they need it (Anthropic). Humans judge two things: tests pass, and it looks right.

Get the bank

One $25 payment, yours forever — up to 10 projects, free updates, no subscription. Your key unlocks the agentic install pack your agent fetches and installs itself. Pay here — you never leave this page.

Updating

Already installed? Paste this to your agent — same key, same install ID, never a new seat:

Update the ai (愛) means love skill bank. You may never have seen this project before — everything you need is in .agents/ailv.json and one open endpoint, no prior context required. There is no update endpoint: updates reuse the install flow with your saved install_id. Read the key from the AILV_KEY environment variable (if it is not set, check the project root .env for it first — if it is nowhere to be found, stop and ask the user to provide it once, then continue with zero further questions) and the install_id from .agents/ailv.json. Never write the key into the project. GET https://aimeanslove.fyi/api/stats (open, no key) and compare its version to the marker. If equal, stop — you are current, nothing to download. Otherwise fetch https://aimeanslove.fyi/api/install?key= with your key appended, write new and changed files, keep the saved install_id, and report success: POST {"version": "VERSION", "install_id": "ID", "installed": [], "kept": []} to https://aimeanslove.fyi/api/install/receipt?key= with your key appended. Same id, new version — updates never count against your seats. Then prune: delete any .agents/skills/<id>/ directory whose id was in the previous marker's installed list but is absent from the new pack — those are orphaned bank files. Never touch kept ids, user-created skills, or non-skill files.

Releases

The bank grows with every release. Each change ships with notes and a feed — read the notes, update on your schedule. Nothing here changes silently under a pinned install.

v1.7.0 — 2026-11-09 — Writing + handover. ailv-app-copy, ailv-code-docs, ailv-code-deliver: web copy, agent docs, and proof-per-task handover. Mobile rules across the UI track, phone-shot gate. Bank at 115 skills.
v1.8.1 — 2026-11-09 — Direct voice. skill openers drop the create-and-install scaffolding and address the installed skill directly; same rules, fewer tokens. Patch: wording only, no behavior change.
v1.9.0 — 2026-11-09 — Output router. ailv-code-output joins the always-on set: match what the user sees to what was done — code proves via deliver, writing shows, research cites. Bank at 116 skills.
v1.10.0 — 2026-11-09 — Links, not diffs. code-shipped output is a clickable file link with line ranges plus test proof — pasted diffs banned, nobody is doing code review.
v1.11.0 — 2026-11-09 — Summarize writing. docs branch stops dumping full text: tight summary plus location, full piece on request only. Grounded in Miller/Cowan chunking, METR debugging overhead, and Stack Overflow verifiability findings.
v1.12.0 — 2026-11-09 — Verdict-first reviews. reviews and summaries lead with the verdict, then three fixed lines — changed, proof, remains. Per BLUF, NNG inverted pyramid, and Lost in the Middle.
v1.14.0 — 2026-11-09 — Honest blocked and ack. blocked states carry the why plus a yes/no or allow-once choice; confirmations go silent when the result is visible. Bank at 116 skills.
v1.13.0 — 2026-11-09 — 60-word reviews. reviews cap at 60 words — our rule, not a research finding. The doc is linked, so nothing is lost by staying short.
v1.8.0 — 2026-11-09 — Helpers + examples. every skill gains scripts/helper.py (runnable, stdlib-only) and examples/output.md (real run output), referenced from frontmatter so prompts stay lean. Ships in both zips; ailv.json gains a layout key.
v1.6.0 — 2026-11-08 — Sized runs. ailv-verify-scope: full suite for behavior, fast path for words — doubt runs everything. Bank at 112 skills.
v1.5.0 — 2026-11-08 — Legal leaves. ailv-app-legal becomes a router over four leaves: pages, card scope, tax, mail consent. Privacy + terms pages published. Bank at 111 skills.
v1.4.0 — 2026-11-08 — Sales law. ailv-app-legal: privacy, terms, refunds, SAQ A, per-region tax, transactional mail — any product, any buyer. Bank at 107 skills.
v1.3.0 — 2026-11-08 — Review gate. ailv-verify-review: post-coding diff review that re-checks matched skills and the philosophy. Bank at 106 skills.
v1.2.1 — 2026-11-07 — Typed failures. code-shape: catch narrow, retry only the transient, log-or-surface the rest. No new skills; bank stays at 105.
v1.2.0 — 2026-11-07 — Namespaced bank. ailv- prefix on all 105 skills: map references always resolve to bank content, never to a same-named user file.
v1.1.0 — 2026-11-07 — Dialog button order. ui-buttons: primary-last order in dialogs (Apple HIG, Material end-aligned confirm). Bank at 105 skills.
v1.0.0 — 2026-11-07 — Signed, leased, measured. 104 skills, every one signed. Lease claims, canary comparator, human-first safety, measured proof section.