Issues

ml: run 6 flags 57% of real documents

stylusnexus/agent-armor#294 · opened

on Oct 8, 2026

ml: build a real-world honest eval set and report detection at low false-flag rates

stylusnexus/agent-armor#276 · opened

on Oct 7, 2026

ml: cut over-defense on short benign text with attack words (NotInject)

stylusnexus/agent-armor#275 · opened

on Oct 7, 2026

1

fix(ml): long documents are scanned only up to token 510

stylusnexus/agent-armor#274 · opened

on Oct 7, 2026

docs: cite OMNI-LEAK, OWASP and public advisory feeds; add an OWASP coverage table

stylusnexus/agent-armor#266 · opened

on Oct 7, 2026

fix(patterns): reduce the 11 accepted false positives (kfp-001 to kfp-011)

stylusnexus/agent-armor#254 · opened

on Oct 6, 2026

fix(patterns): rw-009 fake policy update raising agent limits scans clean

stylusnexus/agent-armor#253 · opened

on Oct 6, 2026

1

fix(patterns): rw-006 fabricated experience log scans clean

stylusnexus/agent-armor#251 · opened

on Oct 6, 2026

1

design: multilingual detection with opt-in language packs

stylusnexus/agent-armor#245 · opened

on Oct 6, 2026

Help wanted: try to break the reference-image scan with your own markdown

stylusnexus/agent-armor#232 · opened

on Oct 5, 2026

fix(ml): ship a LICENSE file in the ML package

stylusnexus/agent-armor#189 · opened

on Oct 4, 2026

chore(repo): turn on secret scanning, push protection, private reporting, CodeQL, tag rules; pin Actions (needs a repo admin)

stylusnexus/agent-armor#188 · opened

on Oct 4, 2026

ci(release): run checks on release PRs (needs a repo admin)

stylusnexus/agent-armor#187 · opened

on Oct 4, 2026

fix(ml): bound the tokenizer cost and checksum every model file

stylusnexus/agent-armor#186 · opened

on Oct 4, 2026

fix(cli): gate and hook bugs (--fail-on none, missing files, symlinks, SARIF)

stylusnexus/agent-armor#184 · opened

on Oct 4, 2026

fix(core): validate remote pattern databases before loading them

stylusnexus/agent-armor#181 · opened

on Oct 4, 2026

feat(detectors): decode and rescan encoded payloads (base64, hex, rot13)

stylusnexus/agent-armor#179 · opened

on Oct 4, 2026

fix(patterns): exfiltration instruction check reads only the matched words

stylusnexus/agent-armor#178 · opened

on Oct 4, 2026

1

fix(core): report a degraded scan and validate strictness instead of a silent clean

stylusnexus/agent-armor#177 · opened

on Oct 4, 2026

Detector: dependency-install risk in scanned content (lockfile bypass, lifecycle scripts)

stylusnexus/agent-armor#107 · opened

on Aug 6, 2026

1

feat(action-gate): package-install policy — release-age floor and pin enforcement

stylusnexus/agent-armor#106 · opened

on Aug 6, 2026

Detector: cross-environment persistence and logging suppression

stylusnexus/agent-armor#99 · opened

on Aug 6, 2026

Detector: C2 / drop-service channel endpoints in scanned content

stylusnexus/agent-armor#98 · opened

on Aug 6, 2026

feat(action-gate): policy ergonomics and hardening (strict rules, tool globs, aggregate reasons, baseDir)

stylusnexus/agent-armor#68 · opened

on Jul 8, 2026

1

Eval-gated external-feed ingestion (patterns + ML training data)

stylusnexus/agent-armor#40 · opened

on Jun 12, 2026

1

Navigation Menu

Repository navigation

Issues

ml: run 6 flags 57% of real documents

ml: build a real-world honest eval set and report detection at low false-flag rates

ml: cut over-defense on short benign text with attack words (NotInject)

fix(ml): long documents are scanned only up to token 510

docs: cite OMNI-LEAK, OWASP and public advisory feeds; add an OWASP coverage table

fix(patterns): reduce the 11 accepted false positives (kfp-001 to kfp-011)

fix(patterns): rw-009 fake policy update raising agent limits scans clean

fix(patterns): rw-006 fabricated experience log scans clean

design: multilingual detection with opt-in language packs

Help wanted: try to break the reference-image scan with your own markdown

fix(ml): ship a LICENSE file in the ML package

chore(repo): turn on secret scanning, push protection, private reporting, CodeQL, tag rules; pin Actions (needs a repo admin)

ci(release): run checks on release PRs (needs a repo admin)

fix(ml): bound the tokenizer cost and checksum every model file

fix(cli): gate and hook bugs (--fail-on none, missing files, symlinks, SARIF)

fix(core): validate remote pattern databases before loading them

feat(detectors): decode and rescan encoded payloads (base64, hex, rot13)

fix(patterns): exfiltration instruction check reads only the matched words

fix(core): report a degraded scan and validate strictness instead of a silent clean

Detector: dependency-install risk in scanned content (lockfile bypass, lifecycle scripts)

feat(action-gate): package-install policy — release-age floor and pin enforcement

Detector: cross-environment persistence and logging suppression

Detector: C2 / drop-service channel endpoints in scanned content

feat(action-gate): policy ergonomics and hardening (strict rules, tool globs, aggregate reasons, baseDir)

Eval-gated external-feed ingestion (patterns + ML training data)