원클릭으로 Manus에서 모든 스킬 실행

deep-hunt

스타0

포크0

업데이트2026년 6월 11일 20:08

Interprocedural, hypothesis-driven hunt — finds the multi-file source→sink bugs that pattern-gating and same-file taint miss. Ranks trace anchors (entry points + dangerous sinks), then the deep-hunter agent walks each flow ACROSS files (forward/backward call-graph CLIs) over multiple rounds, confirming propagation and defeating guards, and promotes confirmed cross-file flows into findings.json with the path stored as evidenceGraph. Token-expensive, budget-bounded. Part of the HUNT phase (run via /sweep --deep, or ask for it); leads flow to /verify (panel).

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

allsmog

allsmog/kuzushi-security-plugin

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

Deep hunt (interprocedural hypothesis loop)

The recall lever beyond /deep-scan (which reads whole files) and /taint-analysis (which traces labeled source→sink, sound only with a CPG): walk a flow across files by forming a hypothesis at an anchor and following the data hop by hop. This is how the multi-file bugs — input in one file, sink in a third — get caught without a 2 GB CPG.

Prefer the provider-neutral command: KUZUSHI_MODEL="${KUZUSHI_MODEL:-openai-codex:gpt-5.5}" node "<plugin root>/scripts/cmd/deep-hunt-run.mjs" --target "<repo root>". Add --input '{"scopeDir":"<dir>","maxAnchors":24}' to scope/bound it. The runner performs prepare -> configured LLM bridge -> finalize without depending on a Claude subagent runtime.
Manual fallback only when the runner is unavailable: run node "<plugin root>/scripts/cmd/deep-hunt-prepare.mjs" --target "<repo root>", read prepPath, walk each anchor with the listed callees/callers CLIs, write { candidates: [...] } to draftPath, then run the returned assembleCommand.
The finalizer enforces that a finding carries a confirmed cross-file path (≥2 hops, ≥2 files), stores it as the finding's evidenceGraph, and promotes under source deep-hunt. Report the flows found + evidence levels, and note they should go through /verify (panel).

When NOT to use

On a single-file script or pure config/binary target — use /deep-scan, /iac, or /binary-recon. Deep-hunt earns its token cost only when flows cross files.
To confirm/prove an existing finding — that's /verify → /poc / /sanitize-pov. Deep-hunt discovers new interprocedural flows.
Before there's any entry surface to start from — build context (/threat-model, /code-graph) first so the anchors and reachability are populated.

Rationalizations to Reject

"A call to the sink exists, so it's a finding." → A call site is not propagation. Confirm the tainted value reaches the sink and that no guard stops it, or it's a candidate.
"No CPG, so I can't trace across files." → Walk the call graph with callees/callers and read each hop; a confirmed cross-file chain is honest linked evidence. Same-file-only is the exact miss this skill fixes.
"Too many anchors — I'll just check a few and call it done." → Spend the budget, then report the unreached anchors + unanchoredCount. Silent truncation reads as "covered everything."
"It's three files deep, skip it." → That depth is the target; the shallow ones the other producers already found.

이 저장소의 다른 Skills

같은 저장소

deep-scan

allsmog/kuzushi-security-plugin

Whole-file deep reader — finds bugs by READING risk-ranked files in full and reasoning from first principles, not by grepping for known patterns. Catches the long tail (project-specific wrappers, business logic, cross-function flaws) that pattern-gated producers structurally miss. Token-expensive; budget-bounded and risk-ranked. Promotes leads into .kuzushi/findings.json (source "deep-scan").

2026-06-110

taint-analysis

allsmog/kuzushi-security-plugin

IRIS-style source→sink taint hunt. Ranks a typed CWE catalog for the repo, then drives subagents to label dangerous sinks, label sources of user input, run Joern/CodeQL dataflow queries (or same-file linking) to connect them, and triage each flow as finding/candidate/rejected. Promotes verdicts into .kuzushi/findings.json. Benefits from a prebuilt CodeQL DB / Joern CPG but degrades gracefully.

2026-06-110

threat-hunt

allsmog/kuzushi-security-plugin

Adversarial per-threat review (Carlini doctrine). For each threat in .kuzushi/threat-model.json, state attacker capabilities, walk source→sink, attempt to bypass every guard, and assign a verdict with evidence. Promotes findings into .kuzushi/findings.json. Requires /threat-model first.

2026-06-110

partition

allsmog/kuzushi-security-plugin

Split the attack surface into parallel-discovery partitions so fan-out hunters explore different subsystems instead of converging on the same shallow bug. Deterministic grouping of /x-ray entry points by component → .kuzushi/partitions.json, which a hunt coordinator hands to one subagent per partition. Run /x-ray first.

2026-06-100

poc

allsmog/kuzushi-security-plugin

Empirical proof-of-concept for the PoC-ready findings. For each finding /verify marked confirmed-exploitable or inconclusive, synthesize a minimal harness that triggers the bug; a host script then runs it in a sandbox (Docker --network none when present, else a gated local run) and classifies the crash into a proof verdict. Attaches a poc block onto each finding. Requires /verify first.

2026-06-100

benchmark

allsmog/kuzushi-security-plugin

Measure recall / precision / false-proof rate of the pipeline against a ground-truth manifest. Scores either the bundled planted-vulnerability corpus (regression) or a live run's findings.json against a manifest you supply. Deterministic — no agent, no network. Use to prove a change to the producers helps rather than hurts.

2026-06-100

name	deep-hunt
description	Interprocedural, hypothesis-driven hunt — finds the multi-file source→sink bugs that pattern-gating and same-file taint miss. Ranks trace anchors (entry points + dangerous sinks), then the deep-hunter agent walks each flow ACROSS files (forward/backward call-graph CLIs) over multiple rounds, confirming propagation and defeating guards, and promotes confirmed cross-file flows into findings.json with the path stored as evidenceGraph. Token-expensive, budget-bounded. Part of the HUNT phase (run via /sweep --deep, or ask for it); leads flow to /verify (panel).
context	command
runner	scripts/cmd/deep-hunt-run.mjs
user-invocable	false

Deep hunt (interprocedural hypothesis loop)

Prefer the provider-neutral command: KUZUSHI_MODEL="${KUZUSHI_MODEL:-openai-codex:gpt-5.5}" node "<plugin root>/scripts/cmd/deep-hunt-run.mjs" --target "<repo root>". Add --input '{"scopeDir":"<dir>","maxAnchors":24}' to scope/bound it. The runner performs prepare -> configured LLM bridge -> finalize without depending on a Claude subagent runtime.
Manual fallback only when the runner is unavailable: run node "<plugin root>/scripts/cmd/deep-hunt-prepare.mjs" --target "<repo root>", read prepPath, walk each anchor with the listed callees/callers CLIs, write { candidates: [...] } to draftPath, then run the returned assembleCommand.
The finalizer enforces that a finding carries a confirmed cross-file path (≥2 hops, ≥2 files), stores it as the finding's evidenceGraph, and promotes under source deep-hunt. Report the flows found + evidence levels, and note they should go through /verify (panel).

When NOT to use

On a single-file script or pure config/binary target — use /deep-scan, /iac, or /binary-recon. Deep-hunt earns its token cost only when flows cross files.
To confirm/prove an existing finding — that's /verify → /poc / /sanitize-pov. Deep-hunt discovers new interprocedural flows.
Before there's any entry surface to start from — build context (/threat-model, /code-graph) first so the anchors and reachability are populated.

Rationalizations to Reject

"A call to the sink exists, so it's a finding." → A call site is not propagation. Confirm the tainted value reaches the sink and that no guard stops it, or it's a candidate.
"No CPG, so I can't trace across files." → Walk the call graph with callees/callers and read each hop; a confirmed cross-file chain is honest linked evidence. Same-file-only is the exact miss this skill fixes.
"Too many anchors — I'll just check a few and call it done." → Spend the budget, then report the unreached anchors + unanchoredCount. Silent truncation reads as "covered everything."
"It's three files deep, skip it." → That depth is the target; the shallow ones the other producers already found.