Borderline — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Borderline (Plugin) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
Aggregate score unchanged between these scans.
The primary manifest — the file an agent reads to learn what this artifact does.
<p align="center"> <img src="img/Borderline.png" alt="Borderline" /> </p>
A Claude Code plugin that delegates to the Antigravity CLI (agy) the boring, mechanical, low-risk tasks where it is just as reliable as Claude: bulk translations and i18n, trivial style/color changes, repetitive renames and replacements, boilerplate. The idea: a transparent Claude ⇄ agy pipeline, so Claude stays reserved for what needs judgment.
skills/borderline/SKILL.md): the brain. It decides what isdelegable, picks the mode, launches agy and always reviews the result. It auto-activates when it detects a borderline task.
agy with two modes:--text → runs agy from an empty scratch directory and returns text only; Claudeapplies the result. Preferred for translations (inline the source) and small changes.
--edit → runs agy in the repo, reading and writing the named files. For bulk work.Because agy is agentic — it takes the working directory as its workspace and scans it ("reads the README", lists files, opens scripts) before answering, even with a concrete prompt — the real fix is structural: --text runs agy from an empty directory, so there is no workspace to explore and no rate-limit quota wasted on it. Each of those exploratory reads is a request against agy's limit, which is why divagation also exhausts the quota. Always go through delegate.sh.
Long context works regardless of size. The wrapper streams the prompt to agy --print on stdin, never as a -p argument — so a large i18n batch or long document can't blow the shell's ARG_MAX ceiling (the old "Argument list too long" / silent hang on long prompts). For content too large to even fit on delegate.sh's own command line, write it to a temp file and pass --text @/path/to/file (or pipe it and pass -); a 1.3 MB prompt that is impossible via -p goes through cleanly.
The wrapper also bounds agy's silent rate-limit (agy throttles by request frequency and hangs returning empty instead of a 429). agy's own --print-timeout doesn't always fire on this stall, so the wrapper adds a hard wall-clock watchdog (--print-timeout + 30 s) that kills a stuck call and prints borderline: HARD TIMEOUT … — the agent never hangs forever. Don't hammer-retry or run calls concurrently — both make throttling worse; for batch i18n, space calls with BORDERLINE_THROTTLE_MS=3000–5000.
Design decisions (set in this build):
/borderline).agy) installed andon the PATH (agy --help).
transparent and doesn't prompt for confirmation on every call.
# In Claude Code:
/plugin marketplace add BansheeTech/Borderline
/plugin install borderline@borderline-marketplaceThen launch Claude with:
claude --dangerously-skip-permissions"switch the background to dark mode") and the skill will delegate to agy and report back.
/borderline translate locales/en.json to locales/de.json.Borderline/
├── .claude-plugin/
│ ├── plugin.json # plugin manifest
│ └── marketplace.json # to install as a local marketplace
├── skills/borderline/SKILL.md
├── commands/borderline.md
├── scripts/delegate.sh
└── README.md~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.