bama
BAMA: Build Any Mac App. Agent skills to research, clone, design, build, optimize, and test native macOS apps with Swift and SwiftUI.
翻譯準備中
get in loser, we're building god
翻譯準備中
技能Python · SolidityGitHub @wildcat-finance自行安裝營運者未確認
試用過了嗎? 點一下就好,不需要登入。
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
安裝型專案無需部署網域也可驗證。個人儲存庫可透過 GitHub 登入驗證,組織儲存庫可使用上方的儲存庫驗證檔案。
正在載入完整 README。 以下為已儲存的摘錄。
The Shoggoth is a collective of small, specialised agents for crypto research
and development, distributed as Wildcat Labs Skills. They preserve
evidence, investigate protocols, build and test software, review contracts,
measure changes, and hand work on without turning uncertainty into fact.檢視完整 README ↗The Shoggoth is a collective of small, specialised agents for crypto research and development, distributed as Wildcat Labs Skills. They preserve evidence, investigate protocols, build and test software, review contracts, measure changes, and hand work on without turning uncertainty into fact.
It is not one general assistant with a cast list. Each member owns a bounded job, says what it checked, shows the evidence behind the result, and stops where another specialist or a person has to take over. There are 27 governed skills in 18 plugins here, and both numbers are derived from the tree each time this page is checked.
Ask the Atlas for a number. Pick your harness. Finish what you start.
Read How to help build the Shoggoth first. It offers small documentation and specimen routes as well as the fully controlled delivery path. If you would rather the system chose, the Shoggoth Wave Atlas draws one unblocked open issue. Its Codex and Claude Code launchers allocate a job, then open the installed local client with the prompt. The manual prompt returns the allocation as JSON. A launcher does not prove that its harness can edit, sign or publish a local change, so keep whichever harness you picked open for the whole run.
No local harness holds a checked one-click Atlas launcher. The probe recorded every harness below and no client answered there, so the roster states what it found rather than what anybody hoped for:
Each harness carries the exact reason it stopped there in the harness table and in docs/harness-classification.json, which both surfaces are generated from.
Use any repository-valid signing identity for commits. Fiat admits valid signatures rather than a named author, trailer, runtime, or GitHub transport. Publication authority remains separate and must come from the repository or the user's explicit instruction.
A daily job rebuilds CONTRIBUTORS.md and the thanks
below from public history. Delivery records whether a pushed commit's author
address matches a GitHub account. If the address matches no account, the
receipt says so instead of guessing. The remaining publication conditions are
GitHub's rather than the repository's.
Much of this page was written by the thing it describes. Read the evidence, not the prose.
The members below rebuild something real, offline, from records preserved in this repository. Each card binds its skill, its claim and the digest of the demonstration ledger beside that skill. Downgrading, editing or removing that record fails the check rather than quietly leaving the claim standing. Each command writes a report to a path that must not already exist.
Anamnesis admits, curates and projects preserved audit specimens from the producer's own bytes, and the mapper a curation policy declares now selects the implementation that reads each source rather than naming the only one there was. Admitting findings a party outside this repository produced, under a mapper declared per source, is the held job.
python3 scripts/demonstrations.py run --record plugins/anamnesis/skills/anamnesis --report tmp/demo/anamnesis.json
Over the preserved plugins/anamnesis/specimens/pilot/policy.json it reports
2. the committed release verifies: 41 finding(s), 31 round(s), 12 with no findings. It does not establish that the corpus is complete, that any finding
is real, or that any remediation is correct.
Lazarus rebuilds its Aave v4 spoke capture at block 25870892 and proves the receipt-trie relation against the recorded header.
python3 scripts/demonstrations.py run --record plugins/lazarus/skills/lazarus --report tmp/demo/lazarus.json
Over the preserved plugins/lazarus/examples/aave-v4-spoke-v1/manifest.json it
reports fixture_rebuild "identical". It does not establish canonical-chain
finality or provider independence for the Aave v4 capture: transaction hashes
are recorded RPC metadata, not a proved header identity, and nothing here
describes current chain state.
Alexandria checks the committed archive manifests, rebuild records and expected values for both Wildcat mainnet intervals: 16 V1 subjects and 137 V2 subjects. It checks the preserved Aave V3 Ethereum segments the same way. Each manifest binds the externally preserved staging files by byte count and SHA-256.
python3 scripts/demonstrations.py run --record plugins/alexandria/skills/alexandria --report tmp/demo/alexandria.json
Over plugins/alexandria/examples/wildcat-estates-interval-v0/expected.json,
the registered operation reports scope "committed-metadata-only" and
rebuild_performed false. The combined demonstration
provides the complete offline rebuild when both external staging trees are present.
This registered operation checks committed metadata only. It neither reads the external staging archives nor rebuilds a release. The separate build and verify commands require the external staging trees. The twelve Aave staging trees are one archive held outside this repository, and the check does not read it. Provider agreement does not establish completeness, publisher identity or canonical-chain finality. Targeted traces exclude transactions without a matching subject log.
Dokimasia regenerates the wildcat-app-v2 coverage
record, scrutiny record and rendered report, and checks that the three agree.
python3 scripts/demonstrations.py run --record plugins/dokimasia/skills/dokimasia --report tmp/demo/dokimasia.json
Over the preserved
plugins/dokimasia/docs/evidence/wildcat-app-v2.coverage.json it reports
dokimasia demonstrate: check clean; a scrutiny is deterministic, each moved identity names its own cause, an unexplained move is reported, and the committed evidence regenerates. The pinned application checkout and workbook
bytes are not preserved here, so this does not regenerate the scrutiny from its
primary inputs or state that any route or workbook case passed; the rendered
report leaves 59 of 261 scoped items without a disposition.
Install the dependency-closed router through the Agent Skills convention:
npx skills add wildcat-finance/skills-runtime --skill promise-machine
That package is generated from this repository and published hourly to
wildcat-finance/skills-runtime,
so an install can sit up to an hour behind main. The installation
guide covers the non-interactive Codex command,
private organisation distribution, and the first-party licence boundary.
The Promise Machine contract is the shared law between first-party skills. State what a successful operation establishes. Name the evidence behind it. Carry the caveats forward when another skill consumes the result. Refuse the dependent action when the evidence is missing or stale, and leave inspection, repair, rerun and rollback available. It checks those relationships and the repository's structure; it does not certify that a domain claim is true.
Futureproofing the Shoggoth is the full catalogue: every member, what it ships today, what evidence is missing, and what it could become with help. 18 domain agents and 9 phase agents sit in it, beside the four bounded delivery worker roles and the upstream Pashov security suite, which ships unchanged. Hexaemeron holds the controlled delivery loop; naming a coding task does not start it.
pyproject.toml declares the supported Python minor, and
.python-version pins the exact interpreter every
Python-backed skill and repository check uses.
ADR-038
records the one-interpreter rule and
ADR-042 the
current transition. Member pages list the rest: Git, GitHub CLI, Foundry, a
Solidity compiler or container runtime, and an archive Ethereum endpoint are
the common ones. Verification and replay are often offline even where capture
is not. AGENTS.md is the file an agent runtime reads.
"Shoggoth" may mean the member currently speaking or the collective as a whole; context decides which. The identity contract records the exact addressing, portrait, authorship and host-provenance rules. A name never activates a skill or widens anyone's authority.
Wildcat Labs first-party work in this repository is licensed under Apache-2.0. The vendored Pashov skills are excluded from that grant and keep their upstream MIT licence and notices.
Thanks to @kethcode, @radup1337, @MunamWasi and @clawdina.