| name | offensive-macos-poc-authoring |
| description | Use when a confirmed candidate or chain-discovery hypothesis is ready for a proof-of-concept harness. The skill guides primitive selection, lab-state preparation, harness scaffolding, reliability capture, and evidence linking back to Scriptorium without losing the discipline that separates a proof from a writeup. Fires on "PoC authoring", "write a PoC", "harness this primitive", "PoC scaffolding", "stand up a chain harness", and "reliability pass". |
| folder | offensive-macos-poc-authoring |
| source | skillz-wave5 |
| trigger_phrases | ["PoC authoring","write a PoC","harness this primitive","PoC scaffolding","stand up a chain harness","reliability pass"] |
PoC Authoring
Channel boundary: REPO_MODE=analysis. PoC code, harnesses, and chain
artifacts produced by this workflow live exclusively in the project clone's
gitignored pocs/ (or chains/) tree. They are not committed to the
reusable station template. Operational tradecraft (persistence, deployment,
evasion, payload delivery) is out of scope; the deliverable is a
reproducible proof in a disposable lab VM.
When To Use
- A candidate has reachability and impact evidence and the operator wants to
prove it end-to-end.
- A chain-discovery row has graduated past
theoretical and the next
experiment is harness work, not more static reading.
- An existing PoC needs a reliability pass — race-window measurement,
prerequisite recording, or version-drift verification.
- The operator is switching Cursor models for harder generation tasks and
needs a clean workspace, scaffolded inputs, and clear evidence linkage.
Lab Topology — Where To Run This
PoC authoring is split between workstation-side scaffolding/text editing and
lab-VM-side harness execution. Workstation actions stay inside the project
clone's gitignored work areas; lab-VM actions are append-logged to
VM_ACTIONS.md.
| Step | Where it runs | How |
|---|
| Select primitive / chain | Workstation | Read CORPUS Exploitability And Chainability + INDEX |
| Pre-flight lab safety | Workstation | Confirm LAB_SAFETY.md lab_disposable: true; create snapshot if R13 applies |
| Scaffold PoC directory | Workstation | Copy templates/poc/README.md.template into pocs/<target-id>/<id>/README.md; create harness sub-directories |
| Author harness | Workstation (model selection chosen for the task) | Cursor edits inside pocs/<target-id>/<id>/ |
| Sync to lab VM | Workstation | scripts/rsync-to-vm.sh --record <target-id> (via existing sync script) or operator-driven scp |
| Run harness | Lab VM | Operator-approved command; append a row to VM_ACTIONS.md |
| Reliability pass | Lab VM | Multiple harness runs; record hit rate, race window, retry strategy |
| Capture evidence | Workstation | Pull harness output into findings/analysis/; link from PoC README |
| Update CORPUS PoC Tracking | Workstation | Edit CORPUS.md PoC Tracking row |
| Anchor in Scriptorium | Workstation | Append a Scriptorium row referencing the PoC ID and Candidate / Chain ID |
If dynamic confirmation needs Ghidra-anchored breakpoints, hand off to
Skills/offensive-macos-gatehouse-ghidra-lldb/SKILL.md for the LLDB anchor
flow.
Workflow
- Pre-flight check:
- Read
LAB_SAFETY.md. If lab_disposable: false, stop and ask the
operator for explicit approval before any dynamic action (R14
standard).
- Read CORPUS Exploitability And Chainability for the candidate or
chain row. The Reachability and Reliability Notes columns drive
harness shape; the Next Experiment column is the entry point.
- Confirm a snapshot strategy in
LAB_SAFETY.md and (if practical)
take a snapshot before any high-disruption action (R13).
- Assign a stable PoC ID:
- Format:
POC-<NNN> zero-padded, sequential within the project.
- Reserve the ID by adding a row to CORPUS PoC Tracking with
Status: scaffolding.
- Scaffold the PoC directory:
- Layout:
pocs/<target-id>/<POC-ID>/ (pocs/, chains/, and
poc/ are gitignored from the station template).
- Copy
templates/poc/README.md.template into the directory and
fill in: target id, candidate / chain id, primitive description,
prerequisites, lab state required, expected impact, risk.
- Create
harness/ for the active code, runs/ for run records,
and evidence/ for artifacts the harness produces.
- Author the harness:
- Start with the smallest harness that exercises the next experiment
in the chain row. Avoid stitching the entire chain in the first
pass.
- Keep the harness checked-in only in the gitignored
pocs/
subtree. The station template never receives PoC code.
- Document any prerequisite the harness assumes (TCC grants,
snapshot state, helper installation) in the README.
- Run on the lab VM:
- Append a
VM_ACTIONS.md row before each run with action label
mcp-tool or other, the command, the outcome (succeeded,
failed, crashed, aborted), and the Snapshot Before value.
- Capture stdout, stderr, log excerpts, and any artifact in
pocs/<target-id>/<POC-ID>/runs/<UTC-timestamp>/.
- Reliability pass:
- Run the harness multiple times. Record hit rate, average latency,
race-window observations, and failure modes.
- If reliability is below the threshold the operator cares about,
update CORPUS Reliability Notes and either iterate or close.
- Update state:
- Set CORPUS PoC Tracking
Status to one of scaffolding,
harness-built, reproducing, reliable, closed.
- Append a Scriptorium row with the PoC ID, Candidate / Chain ID,
evidence path, and current claim.
- Append a Chronicle row when the PoC's status changes meaningfully.
- Close the loop:
- When the PoC is
reliable, hand off to
Skills/offensive-macos-submission-packet/SKILL.md for vendor
disclosure or internal reporting.
- When the PoC is
closed (false positive, version-bound, etc.),
record the rationale in the PoC README and the Chronicle.
PoC Status Values
scaffolding: directory created, README authored, no harness yet.
harness-built: harness exists, has not been run on the lab VM.
reproducing: harness has run, primitive triggers, reliability not
yet measured.
reliable: harness reproduces with operator-acceptable reliability.
closed: PoC is no longer worth pursuing; rationale in the README.
These match the values the CORPUS PoC Tracking column expects.
Evidence Discipline
A PoC counts as "real" only when it carries:
- A target ID and a candidate / chain ID it proves.
- Lab VM identity (OS build, SIP state, arch, snapshot reference).
- The harness code committed in the project clone (gitignored from
the template).
- At least one reproducing run record under
runs/.
- A reliability note (single-run, race-bound with measured hit rate,
deterministic, etc.).
- A Scriptorium anchor and a CORPUS PoC Tracking row.
Without those pieces the PoC stays in scaffolding or harness-built
status and does not count for reporting.
False-Positive Traps
- A harness that runs once and produces the expected log line proves
almost nothing. Reliability is part of the proof.
- A harness that requires post-snapshot state the snapshot does not
capture is brittle; record the prerequisite explicitly or it will
not survive the next reset.
- A chain harness that stitches three primitives in one pass usually
hides which primitive failed. Author each primitive's harness first,
then a stitcher.
- A PoC whose reliability depends on a specific VM build is still a
PoC, but the build is part of the proof — record it.
- A "successful" harness without a captured run record is not evidence;
rerun and capture.
Pitfalls
- Do not commit PoC code, harnesses, run records, or chain artifacts
to the station template repo. They live in the project clone's
gitignored
pocs/<target-id>/<POC-ID>/ directory.
- Do not skip the
VM_ACTIONS.md append for harness runs even when
the lab VM is disposable; the audit log is what makes the PoC
reproducible by another operator.
- Switching Cursor model selection mid-PoC is fine, but do not
re-author what is already working; treat the current harness as
authoritative and iterate on it.
- The submission packet is downstream of
reliable; do not promote a
reproducing PoC to a vendor report until reliability is recorded.
See Also
templates/findings-repo/POC_SCAFFOLDING.md
templates/findings-repo/templates/poc/README.md.template
Skills/offensive-macos-chain-discovery/SKILL.md
Skills/offensive-macos-gatehouse-ghidra-lldb/SKILL.md
Skills/offensive-macos-scriptorium-evidence/SKILL.md
Skills/offensive-macos-submission-packet/SKILL.md
templates/findings-repo/LAB_SAFETY.md
templates/findings-repo/VM_ACTIONS.md