fix(ENG-1961): defender plugin — payload floor, cue threshold, bump to 0.8.2

[hiskudin]

Summary

Four fixes to stackone-defender surfaced during Tier-2 false-positive root-causing.

1. PAYLOAD_SKIP_BELOW_BYTES: 500 → 32 (2628b8d) — the pre-daemon size gate silently dropped scans on any tool output under 500 bytes, including short overt injections that fit in ~120B. Verified via hook probe: same attack text flagged at 720B (with padding), silent at 114B. The gate exists to save an IPC round-trip on payloads defender itself would skip (per-string floor: 10 chars); 500B was a 50× overshoot of that rationale. New value covers the JSON wrapper (~20B) plus safety margin. Regression guard in tests/payload-skip-floor.test.mjs.

2. maxSentence display: 80 → 300 chars (17d24bc) — the offending chunk in a real Tier-2 FP was 331 chars long; the 80-char truncation hid what actually tripped the classifier. Diagnostic-only, no security impact.

3. @stackone/defender: ~0.7.0 → 0.8.2 (exact pin) (ef1d356) — picks up the shipped Tier-2 detect-and-gate work (decorative-run normalization, NFKC unicode fold, token-degeneracy guard). Was invisible to the shipped plugin because the previous stale lockfile held resolved version at 0.7.0. Also drops the tilde per the standing "exact pin, no ranges" convention, and adds the previously untracked package-lock.json so lockfile drift can't recur silently.

4. Suspicious cue threshold: 0.3 → 0.64 (d47054e) — plugin-side advisory tier fired below the library's own 0.5 medium-risk bar; on out-of-sample benchmarks the 0.3–0.5 band contributed disproportionate advisory noise per unit of true-positive signal. Aligns the advisory threshold with the block threshold: block or silent, no middle warn band. Matches the SKILL's silent-review posture.

Test plan

• npm test — 13/13 (12 QA fixtures + ENG-1961 regression guard) on defender 0.8.2
• Live-session testing in Claude Code — verify FP rate drops on real tool output, no regressions on realistic-injection fixtures
• Peer hook-level probe battery + benchmark FPR sweep on the deployed build

Follow-ups (not in this PR)

• Advisory else-branch in scan-tool-result.mjs is now effectively unreachable given the aligned threshold — left in place as an explicit no-op; can drop if the block-or-silent posture holds.

---

Summary by cubic

Reduces false positives and silent misses in the Defender plugin by scanning small tool outputs, aligning cues with blocking, and pinning dependencies.

• Scan gate: lowers PAYLOAD_SKIP_BELOW_BYTES from 500 to 32 so sub-500B tool outputs are scanned; adds a constant-ceiling regression test and a below-floor hook smoke test; adds a long benign-text fixture to the QA suite; README reflects the ~32B floor.
• Diagnostics: expands maxSentence display from 80 to 300 chars to aid triage.
• Dependencies: pins @stackone/defender to 0.8.2 and adds package-lock.json; adds overrides for tar, protobufjs, sharp, adm-zip, and nanoid to clear audit issues.
• Cues: aligns advisory with the block threshold (0.64) and removes the inert Suspicious branch; behavior is now block-or-silent.

^{Written for commit 1a9c61b. Summary will update on new commits.}

[Copilot]

Pull request overview

This PR updates the stackone-defender security plugin to reduce false negatives/noise during Tier-2 false-positive root cause analysis by lowering the payload-size skip floor, updating the Tier-2 library dependency, and aligning cue thresholds and diagnostics with current Tier-2 behavior/config.

Changes:

• Lowered the hook-side payload byte gate (PAYLOAD_SKIP_BELOW_BYTES) and added a regression test to prevent it from drifting upward.
• Increased maxSentence diagnostic visibility and raised the “suspicious cue” threshold to align with configured Tier-2 thresholds.
• Pinned @stackone/defender to 0.8.2 and added a lockfile to prevent silent resolution drift.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 1 comment.

File	Description
plugins/security/stackone-defender/tests/payload-skip-floor.test.mjs	Adds regression coverage to prevent raising the payload-skip floor back to overly large values.
plugins/security/stackone-defender/scripts/scan-tool-result.mjs	Lowers IPC skip floor, expands diagnostic maxSentence display, and adjusts suspicious-cue threshold.
plugins/security/stackone-defender/package.json	Pins @stackone/defender to 0.8.2.
plugins/security/stackone-defender/package-lock.json	Adds lockfile to ensure deterministic dependency resolution.

Files not reviewed (1)

• plugins/security/stackone-defender/package-lock.json: Generated file

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[cubic-dev-ai]

All reported issues were addressed across 4 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed across 5 files (changes from recent commits).

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed across 1 file (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>

[cubic-dev-ai]

0 issues found across 2 files (changes from recent commits).

<sub>Requires human review: Lowers scan skip threshold, increases diagnostic truncation, pins defender dependency to 0.8.2, removes dead advisory branch. Requires human verification of dependency update and advisory removal.

Re-trigger cubic</sub>