firmware-exploitability-orchestrator

Use when a candidate has plausible reachability or runtime evidence and same-session follow-up work needs controlled probing, debugging, root cause, or exploitability modeling.

SKILL.md

GitHub

Firmware Exploitability Orchestrator

Overview

Advance a candidate from reachability and runtime evidence toward exploitability assessment. This skill is part of the discovery-first firmware research suite for authorized binary/service exploitability assessment. It preserves blocker-first workflow, compact local evidence, and provenance-first claims.

Routing

role: orchestrator
purpose: Advance a candidate from reachability and runtime evidence toward exploitability assessment.
when_to_use:
  - Use when a candidate has plausible reachability or runtime evidence and same-session follow-up work needs controlled probing, debugging, root cause, or exploitability modeling.
when_not_to_use:
  - Do not probe without hypothesis, input vector, expected observation, negative controls, stop condition, and safety boundary.
required_inputs:
  - hypothesis_ledger.json
  - input_vectors.json
  - service_state_map.json
  - runtime_observation.json or blocker
produces:
  - next_skill_decision.json
  - probe_plan.json
  - debug handoff
  - exploitability_assessment request
completion_gate:
  - The next interaction or domain skill is justified by the highest-risk missing evidence level.
reentry_triggers:
  - Re-enter after probing, sink hits, crashes, debugging results, or exploitability model changes.
handoff_hints:
  - Route to safe probing, runtime observation, debugging, crash root cause, or exploitability modeling.
misuse_signals:
  - The skill is being misused if it creates payloads without authorization boundaries or stop conditions.

Required Superpowers

  • Use superpowers:systematic-debugging when validation fails unexpectedly.
  • Use superpowers:verification-before-completion before claiming an artifact is valid.

Inputs

  • hypothesis_ledger.json
  • input_vectors.json
  • service_state_map.json
  • runtime_observation.json or blocker

Workflow

  1. Check candidate evidence level and missing exploitability fields.
  2. Prefer observation before probing when reachability is uncertain.
  3. Require debugging only for crashes, suspicious effects, sink hits, or root-cause questions.
  4. Block exploitability claims until primitive, impact, constraints, auth requirement, and reproducibility exist.

Always use firmware-artifact-contract before trusting shared artifacts. Every JSON output must include schema_version, generated_at, generated_by, source_inputs, warnings, and errors. Use missing_tool warnings or blockers when required tooling is unavailable. Static evidence, sandbox_generated evidence, Qiling-only output, and public writeups are not runtime truth; keep behavior_claim_allowed=false unless observed_runtime_qemu, observed_runtime_live_hook, observed_runtime_live_debugger, or verified evidence proves the target workload was observed.

Session Continuation

Treat same-session follow-up prompts about the current firmware artifact as continuation signals. Resume from the latest artifact index, hypothesis ledger, next_skill_decision.json, or blockers.json instead of waiting for the user to invoke the skill again explicitly.

Loop Guard

Do not repeat the same failing action, validation, probe, or handoff when inputs and artifacts have not changed. If the same blocker appears twice, write or update blockers.json with the repeated blocker, stop and ask the user for the missing input or decision. Prefer a lower-cost evidence-gathering step only when it changes the artifact state.

Outputs

  • controlled effect
  • root-cause question
  • primitive
  • impact
  • constraints

Verification Gate

Do not mark this skill complete until outputs satisfy the routing completion gate and the artifact contract. Evidence levels must progress from E0 to E8 without skipping runtime or duplicate-check gates. verification-before-completion is required before claiming a result is valid.

Safety

Operate only on firmware and runtimes the user is authorized to test. Keep destructive probes disabled by default; record runtime modifications; prefer local evidence and redacted summaries; use exact dates and versions in vulnerability reports; do not install tools without explicit user approval. Do not commit firmware, secrets, proprietary dumps, raw traces, or private vendor material. Do not mass target devices, do not scan public IPs, do not build a weaponized exploit generator, and do not invent CVE IDs.

Common Mistakes

  • Using probing as discovery without controls.
  • Calling debugging to compensate for missing inventory.
  • Claiming exploitability below E6.