Releases: DouglasVulcano/zeroth-ai
Releases · DouglasVulcano/zeroth-ai
Release list
v1.5.0 - Zeroth
Changed
- Renamed the project to Zeroth. The plugin id is now
zeroth(install with
/plugin install zeroth), the skill lives atskills/zeroth/, the slash command is/zeroth, the
spec iszeroth.md, and the repository isDouglasVulcano/zeroth-ai(site at
douglasvulcano.github.io/zeroth-ai/). Update any marketplace add toDouglasVulcano/zeroth-ai.
Existing installs should reinstall under the new name (/plugin install zeroth).
v1.4.0
Added
- Scaffolder branch-protection guidance is now ruleset-aware. It prints both the classic
branch-protection command and a Ruleset command (POST /repos/{o}/{r}/rulesets), and a new
--rulesetflag makes--protectcreate a ruleset instead of classic protection. Both paths warn
thatenforce_adminsis not a status-check context (it is theenforce_adminsfield in the classic
API and thebypass_actorslist in a ruleset). SECURITY.md documents the ruleset equivalent. This
closes the gap where hand-adapting the classic JSON into a ruleset could hang a PR at
"Expected - Waiting for status to be reported".
v1.3.2
Security
- Pinned the repo's own CI action (
actions/checkout) to a full commit SHA with a version comment,
and added.github/dependabot.yml(grouped, weekly) to keep it current. SECURITY.md now states the
posture accurately: this repo pins to SHA; scaffolded templates default to major-version tags
(which receive patches) and can be pinned for the strictest posture.
v1.3.1
Added
- README (EN and PT-BR): a single "golden path" quickstart (scaffold, fill AGENTS.md, set
CODEOWNERS, arm branch protection) plus a "this repo runs on its own standard" note pointing to
the end-to-end fixtures. The branch-protection step flags the Ruleset gotcha (enforce_adminsis
not a status-check context).
v1.3.0
Added
- Scaffolder now arms the authoritative gate.
scaffold.shprints a branch-protection command
(gh api ... /branches/<branch>/protection) personalized from the target'soriginremote and
default branch, and an opt-in--protectflag applies it (needs gh + a repo-admin token; honors
--dry-run, confirms on a TTY). It also warns when the generatedverify.ymlis the generic
placeholder, so you do not require a status check that verifies nothing. scripts/verify.shcovers the new scaffolder output (branch-protection guidance;--protect
respects--dry-run).
v1.2.2
Fixed
- Quickstart: the global-skill install snippet referenced the wrong directory after
git clone
(cd ai-engineering-standardsshould becd ai_engineering_standards), in README (EN and PT-BR). - CHANGELOG accuracy: the v1.2.0 note overstated eval automation. CI validates the eval-suite
structure viaverify.sh; the evals themselves run manually (claude plugin eval).
v1.2.1
Security
- Hardened the Bash guard: it now also catches shell-wrapper evasion (
bash -c,sh -c,eval),
piping a download into a shell (curl ... | sh), and writing to secret files via redirection
(> .env,id_rsa,*.pem,.git-credentials,.npmrc). Still fail-open, still a guardrail and
not a sandbox. - Scaffolded Node CI installs with
--ignore-scripts; both CI templates document version and action
SHA pinning. - Added
.github/CODEOWNERSand a "Threat model and trust boundary" plus "Repository hardening"
section to SECURITY.md (branch protection, required Code Owner review, tagged releases).