Skip to content
Back to skills

Venus Packet Preflight

ASecurity

Implementer pre-flight for a Venus wave packet, the checklist and the tools/check_packet.py script that catch the defect classes reviewers rejected in every Stage A wave (bare assert() in Release, engine getenv levers, vacuous WILL_FAIL controls, unmeasured numbers, 70-line functions, comment terminators). Run BEFORE handing off any packet; paste its block into the handoff.

  • 9 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 27, 2026
ai-agentspythongobashexpressgitapi

Works with

  • api

Security analysis

A100/100

Scanned September 27, 2026

npx -y skills add Ugbot/ai-grind --skill venus-packet-preflight --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Venus Packet Preflight?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Venus Packet Preflight
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ugbot-venus-packet-preflight/badge)](https://www.skillsdirectory.com/skills/ugbot-venus-packet-preflight)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: venus-packet-preflight
description: Implementer pre-flight for a Venus wave packet, the checklist and the tools/check_packet.py script that catch the defect classes reviewers rejected in every Stage A wave (bare assert() in Release, engine getenv levers, vacuous WILL_FAIL controls, unmeasured numbers, 70-line functions, comment terminators). Run BEFORE handing off any packet; paste its block into the handoff.
---

# Why this exists

Doc 34 (`docs/design/game/34-stage-a-retrospective-20260908.md`): 141 reviewer
findings over four waves; three classes recurred in EVERY wave after being
called out, bare `assert()` (16), non-discriminating WILL_FAIL controls (13),
numbers written before measurement (11). Implementers fixed only the lines a
reviewer enumerated. This skill makes the checks mechanical.

# Run it

```bash
cd C:/code/Venus
python tools/check_packet.py --base <wave base SHA> --handoff
# restrict to your packet's files:
python tools/check_packet.py --base <sha> --files src/physics/x.c tests/x_test.c --handoff
```

Exit 1 = ERROR class present (E1 bare assert in src/, E2 getenv in an engine
TU, E3 WILL_FAIL without a PASS_REGULAR_EXPRESSION sentinel). Fix, re-run,
paste the `PRE-FLIGHT` block into the handoff's `notes`.

# The checklist (each item is a grep or a printed line in the handoff)

1. **Asserts**: `assert(` count in your diff under `src/` = 0. Every new
   function has >= 2 `VENUS_ASSERT` including negative space; none is a
   store/readback or `x == true || x == false` tautology. Data conditions
   (world geometry, input, I/O) get error paths, not asserts.
2. **Levers**: no new `getenv()` in engine TUs. In-process test levers are
   explicit setters called from the test's `main()` BEFORE any world/server
   exists (precedent `venus_physics_set_force_serial`,
   `venus_mesh_active_mask_force`, `job_system_set_legacy_retry`,
   `wyrd_net_server_set_phase_tally_enabled`). MSVC trap: `uv_os_setenv` /
   `SetEnvironmentVariable` never refreshes the CRT copy `getenv()` reads,
   an env var set in-process is invisible to the engine. ctest `ENVIRONMENT`
   is fine (it is in the initial process block).
3. **Cross-thread globals**: `_Atomic` with a live
   `VENUS_ASSERT(atomic_is_lock_free(&g))`, release store / acquire load; or
   the handoff proves single-thread-before-start.
4. **Falsifiers**: one printed sentinel line `NAME key=value ...`, registered
   with `PASS_REGULAR_EXPRESSION` on that sentinel, NO `WILL_FAIL`, no line
   numbers in the regex. See `venus-falsifier`.
5. **Numbers**: a number in a handoff, MAP note or comment is copied from a
   named log line (`scratchpad/<wave>/... :line`) or written as `UNMEASURED`.
   Never a cross-run ratio. Never a triage figure after a build measured it.
6. **Findings ledger**: `findings_addressed` + `unfinished` together list
   EVERY finding id of the review round, by the reviewer's numbering.
   Dropping one silently is a major on the next review.
7. **Lengths**: every touched function's post-length; a touched > 70-line
   function is split or the handoff says why not. A touched > 1500-line file
   ends shorter or is split. New features go in NEW TUs.
8. **Fixture conventions**: read the API header before writing a fixture and
   quote the convention in the leg's comment. Known ones: capsule query
   `origin` is the CENTRE (`venus_pquery.c`, `c1 = origin - half_height`);
   the server drains hellos then steps in the SAME tick; admission traces
   have fixed capacities; `system()` on cmd.exe strips quotes.
9. **Compile proxy** (you may not build): brace/paren balance on every edited
   file; grep `*/` inside doc comments in edited headers; every called symbol
   opened in its header and the signature checked; re-read the whole diff.
10. **Ownership**: touch ONLY the packet's files; never `git add/commit/stash/
    checkout/reset`; MAP notes go to `src/<area>/MAP.waveN-<packet>.md`.

# Handoff shape

Production consumer (`file:line`) · changed/new files · gates with their
negative control and how it is switched · targets to build · findings
addressed / unfinished (all ids) · owner-only questions · the PRE-FLIGHT
block. A claim about a gate result you did not run is "reasoned, unrun".

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…