Skip to content

Releases: alisas-cell/de-starter

Release list

de-starter v0.1.2

de-starter v0.1.2 Latest
Latest

Choose a tag to compare

View all tags
@alisas-cell alisas-cell released this 23 Jul 07:59
v0.1.2
65b99eb

What’s new

  • Added a fully synthetic public demo lab with sentinel-owned prepare, inventory, stale-preview exercise, post-apply checks, and safe reset commands.
  • Added a staged five-minute walkthrough that preserves both human approval gates and never auto-extracts or submits a preview token.
  • Added wrong-token and stale-preview refusal tests proving zero partial approved edits, plus protected P0/P1 and ordinary-empty-directory invariants on the success path.
  • Added an explicit non-zero-risk safety boundary, a redacted 16:9 evidence card, and updated Chinese self-media materials.
  • Expanded the full regression suite from 195 to 214 tests.

Verification

  • 214/214 full regression tests passed locally in strict ResourceWarning mode.
  • GitHub Actions passed on Python 3.9, 3.11, and 3.13 for both main and the v0.1.2 tag.
  • A fresh-clone public-demo rehearsal passed, including discover, audit, preview, approval, apply, verify, check, and reset.
  • Skill validation, compile, JSON, Markdown-link, image-dimension, and privacy checks passed.

No production Skill behavior changed from v0.1.1.

Assets 2
Loading

de-starter v0.1.1

de-starter v0.1.1

Choose a tag to compare

View all tags
@alisas-cell alisas-cell released this 23 Jul 02:53
v0.1.1
437d3d6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Patch release

v0.1.1 keeps the production Skill behavior from v0.1.0 and closes the two issues found during first-time public release verification.

  • Made the preview-root identity regression portable to Linux filesystems that may immediately reuse an inode.
  • The regression still proves that the approval token tracks preview-root identity and that an explicit keep decision changes the decision hash.
  • Updated actions/checkout and actions/setup-python to the official Node 24 runtimes.

Verification

  • GitHub Actions passed all three jobs on Python 3.9, 3.11, and 3.13.
  • The successful workflow run has no annotations.
  • Local strict verification remains 195/195 full tests and 13/13 CLI lifecycle tests.
  • No production Skill code changed in this patch.

For the complete feature list, safety boundary, real Starter acceptance result, installation steps, and public evidence, see the v0.1.0 release and the repository README.

Loading

de-starter v0.1.0

de-starter v0.1.0

Choose a tag to compare

View all tags
@alisas-cell alisas-cell released this 23 Jul 02:11
v0.1.0
8bd9f42

What this is

de-starter is a safety-first Agent Skill for auditing and removing starter, boilerplate, template, and SaaS-kit residue. It guides an existing coding Agent; it is not a standalone Agent.

Highlights

  • Discovers source identities instead of relying on one hard-coded brand list.
  • Classifies findings as P0–P3 so legal, secrets, production identifiers, demos, and display branding receive different treatment.
  • Requires two approval gates: an audit/decision gate and an exact preview/token gate.
  • Supports protected line-level semantic edits with audited hashes, explicit ranges, and migration/rollback plans for P1 overlap.
  • Audits source-named directories separately and only removes an exact approved empty directory.
  • Keeps preview, byte backups, reverse diff, restore manifest, and recovery evidence outside the target project.
  • Fails closed on drift, symlink races, no-clobber violations, unsupported filesystem primitives, and cross-filesystem apply.

Verification

  • 195/195 full regression tests pass with ResourceWarning treated as an error.
  • 13/13 CLI lifecycle tests pass.
  • Skill structure validation, compile checks, JSON/YAML parsing, Markdown links, privacy scan, and diff checks pass.
  • A fresh public clone, install copy, and complete 195-test run pass.
  • Real Starter acceptance: file findings reduced from 523 to 227; the separately approved empty-directory residue reduced from 1 to 0. Protected P0/LICENSE content remained, no new inherited project test failures were introduced, and the production webpack build passed.

Safety boundary

v0.1 automated commands support macOS/Linux POSIX no-follow primitives. Apply requires the target project and external run directory to be on the same filesystem. Windows is not supported in v0.1. The goal is not “zero keywords”: remaining findings can be legal evidence, compatibility identifiers, or functionality the user explicitly keeps.

Never publish purchased source code, private diffs, credentials, approval tokens, or backup mappings.

Install

git clone https://github.com/alisas-cell/de-starter.git
mkdir -p "$HOME/.agents/skills"
cp -R de-starter/skills/de-starter "$HOME/.agents/skills/de-starter"

See the README, sanitized real-run report, screenshots, and Chinese self-media package in the repository.

Loading