psigho
6ba02ec50d
Detection reported installed npm-global agents (Claude Code, Codex) as
'binary not found on PATH' on Windows, and an authed Hermes desktop as
'NOT AUTHED'. Two root causes:
1. resolvePath() used POSIX-only 'command -v'/'which', and
execFile/spawn('claude', ...) cannot resolve npm's .cmd shims via
PATHEXT -> ENOENT -> installed:false. Add resolveBin() (where.exe on
Windows, preferring .exe > .cmd > .bat) and route detection + both
spawn call sites through the resolved path.
2. Hermes desktop on Windows stores its login under %LOCALAPPDATA%\hermes not ~/.hermes/. Add localAppData() and check both (presence-only).
Windows .cmd shims are launched via a new spawnAgent() helper that invokes
cmd.exe with an argv ARRAY (shell:false) instead of shell:true, so the
untrusted prompt is never re-parsed by cmd.exe -> no command-injection surface.
No new deps; typecheck clean.
60 lines
2.2 KiB
Plaintext
Executable File
60 lines
2.2 KiB
Plaintext
Executable File
node_modules/
|
|
dist/
|
|
.env
|
|
*.tsbuildinfo
|
|
|
|
# --- excluded from the published repo ---
|
|
# raw upstream Cybench corpus (181M third-party content incl. writeups/flags/solutions — do not redistribute)
|
|
bench/cybench/corpus-stage/
|
|
# ephemeral docker staging (regenerated each run)
|
|
bench/cybench/service-stage/
|
|
# per-challenge staging dirs (contain flag.expected answer files; regenerable)
|
|
bench/cybench/challenges/
|
|
# XBOW/XBEN staging (1.3G third-party challenge source + .env flag files; regenerated by xbow-bench.mjs)
|
|
bench/xbow/stage/
|
|
bench/xbow/challenges/
|
|
# robustness fallback dirs + artifact unpacks (regenerable)
|
|
**/*.stale-*
|
|
**/release_unpacked/
|
|
# secrets / creds (belt-and-suspenders)
|
|
.env
|
|
.env.*
|
|
!.env.example
|
|
*.pem
|
|
*.key
|
|
# logs + python caches + os cruft
|
|
*.log
|
|
**/__pycache__/
|
|
.DS_Store
|
|
|
|
# --- coordinated-disclosure HOLD (never publish unpatched vuln detail) ---
|
|
# Structured findings, executed PoCs/reproducers, generated disclosure packages,
|
|
# and reporting guidance stay LOCAL until each bug is fixed/disclosed with its
|
|
# vendor. The *tooling* that makes them (disclosure-gen, verify-finding) is
|
|
# tracked; its *outputs* — and any reproducer code or compiled PoC binary — are not.
|
|
# Entire wild-hunt tree = unreported-vuln hunt output (structured findings, executed
|
|
# PoCs, generated disclosure dossiers, the results/ campaign summaries, GATE-PRECISION).
|
|
# NONE of it ships until each bug is vendor-coordinated. The *tooling* lives in src/ +
|
|
# scripts/; only its *outputs* live here, so ignoring the whole tree is safe (0 tracked).
|
|
bench/wild-hunt/
|
|
docs/disclosures/
|
|
# Held planning docs that reference unreported vulnerabilities (pre-disclosure)
|
|
docs/V6_PLAN.md
|
|
|
|
# decomposition run artifacts (reports may contain unreported vulnerability analysis)
|
|
bench/decomposition-results/
|
|
|
|
# refusal-frontier raw responses can contain model-written toy-exploit text
|
|
bench/refusal-frontier/
|
|
bench/nyu/
|
|
|
|
# local one-off key paste file (Venice etc.) — never commit
|
|
.keys.local
|
|
# bounty platform API credentials — never commit
|
|
.keys.bounty.json
|
|
# sensitive export leak-gate denylist (operator handle + embargoed disclosure
|
|
# targets + personal vault) loaded by the exporter — never commit
|
|
scripts/.export-denylist.local
|
|
server.pid
|
|
server.log
|