bootproof 0.4.0 → 0.4.1

package/README.md

@@ -5,6 +5,10 @@
 
 BootProof answers one question: **did this repository actually boot?** Not "did a command run?" Not "did Docker say containers are up?" Not "did an AI agent say it worked?" BootProof inspects a repo, builds an evidence-based run plan, executes only what it can justify, observes real health, and writes a signed attestation for success or failure. No proof, no green check.
 
+<p align="center">
+  <img src="assets/bootproof-supabase-demo.gif" alt="BootProof verifying a Supabase-style stack: infers the boot path, starts services, observes HTTP health, writes a signed attestation" width="900">
+</p>
+
 ---
 
 ## The Living Receipt: proof that travels

package/dist/proof.d.ts

@@ -1,5 +1,5 @@
 import type { Attestation, AttestationTrust, ObservedStep, RunPlan, FailureClass, HealthEvidence, VerificationMode, ExternalVerificationClassification } from "./types.js";
-export declare const TOOL_ID = "bootproof@0.4.0";
+export declare const TOOL_ID = "bootproof@0.4.1";
 export type { AttestationTrust } from "./types.js";
 export type SignerTrustTier = "invalid" | "self" | "known" | "unknown-foreign";
 export interface SignatureTrustResult {

package/dist/proof.js

@@ -5,7 +5,7 @@ import path from "node:path";
 import { execFileSync } from "node:child_process";
 import { buildExecutionEnv } from "./exec.js";
 import { redactJsonValue } from "./redact.js";
-export const TOOL_ID = "bootproof@0.4.0";
+export const TOOL_ID = "bootproof@0.4.1";
 export function gitInfo(repo) {
     const git = (...args) => {
         try {

package/docs/DISTRIBUTION.md

@@ -0,0 +1,83 @@
+# Distribution Kit v2
+
+The first launch was never seen. HN auto-kills link posts from new accounts silently; Reddit automod filters new accounts before humans see them. The market has not said no — it has not been asked yet. This kit is the re-ask, through channels with no bouncer at the door.
+
+**The one metric for the next 30 days: receipts in the wild that you did not generate.** Target: 10. Everything below exists to move that number.
+
+---
+
+## Part 1 — Recover Hacker News
+
+1. Log out of HN (or set `showdead: yes` in your profile) and view the Show HN post. If it shows [dead], it was auto-killed and zero humans saw it.
+2. Email the moderators. They are responsive, they vouch legitimate posts from new accounts, and they run a second-chance pool for good posts that sank unfairly.
+
+> To: hn@ycombinator.com
+> Subject: New account, Show HN appears auto-killed
+>
+> Hi — I posted a Show HN for my open-source project (link to the post) from a new account and I believe it was caught by the spam filter. It's a runtime verification tool for AI-generated code (github.com/bootproof/bootproof), genuinely my own work, Apache-2.0 licensed. Would appreciate a look — happy to answer anything. Thanks for what you do.
+
+3. Before any repost: (a) the 60-second GIF exists and is the first thing in the README; (b) spend a week leaving genuinely useful comments on other threads so the account has history. Reposting a no-traction story is explicitly allowed.
+4. Repost Tue–Thu, 14:00–16:00 UK (morning US). Title suggestion — note that plain "BootProof" collides in search with a football-boot bag brand, so lead with the claim, not the name:
+   "Show HN: Signed receipts proving my AI agent's code actually runs"
+
+---
+
+## Part 2 — Receipt-First Outreach
+
+The mechanic that no other product can copy: the email CONTAINS the proof, about THEIR repo. Process per target, ~20 minutes:
+
+1. `bootproof up <their-repo> --provider local --unsafe-local --receipt`
+2. Whatever happens is the opening line:
+   - Boots clean → attach Living Receipt: "here's signed proof your repo boots from a fresh clone."
+   - Honest refusal (like Airbyte) → even better story: "my tool refused to pretend it could boot your stack — here's the signed refusal."
+3. Send. Short. No ask beyond "curious what you think." Never attach to a repo's issue tracker unless their contributing docs invite tooling discussion — email or their stated contact route only.
+
+### Template A — maintainer publicly weary of AI slop
+
+> Subject: Signed proof {repo} boots from a fresh clone (41s) — thought of your AI-slop posts
+>
+> Hi {name} — your writing about the burden of AI-generated {reports/PRs} is half the reason I built this. BootProof boots a repo under supervision and signs what it observed; no observed health signal, no green check. I ran it against {repo}: {one-sentence honest result}. The receipt is attached — it's a single HTML file that re-verifies its own signature in your browser, and visibly collapses if you tamper with a byte (there's a button).
+>
+> There's also a CI gate so agent PRs can't merge without an observed boot. If it's useless to you, that's a data point I want too. Either way — thanks for the writing.
+
+### Template B — repo you already tested (Airbyte-class)
+
+> Subject: Your repo, cryptographically refusing to lie about itself
+>
+> Hi — I built a runtime verifier and {repo} was one of my test cases. It {booted clean in Xs / correctly refused: your documented path needs {abctl/kind/helm}, and my tool declines to pretend a bare command is enough}. Signed receipt attached; it verifies itself offline. Sharing because the result is genuinely about your project, not mine. Would value 60 seconds of your reaction.
+
+### Template C — agent-tool authors (Claude Code hooks, agent frameworks)
+
+> Subject: A Stop-hook that makes agents hand over signed proof their work runs
+>
+> Hi — I wrote a hook for {tool}: when the agent claims "done," it boots the workspace under supervision and prints a signed verdict before the human reviews anything. Fails closed. One-liner config attached, receipt from a real run attached. If this fits your examples/integrations page, it's MIT — take it. If it's a bad fit, one line saying why helps me more than silence.
+
+### First ten targets
+
+1. Daniel Stenberg (curl) — has documented at length the burden of AI-generated junk reports. Template A.
+2. Seth Larson (Python security) — has written publicly about AI-generated security reports. Template A.
+3. The Airbyte team — you already have the honest-refusal receipt. Template B.
+4–6. Every other repo in your test fixtures with a real maintainer. Template B — these are pre-written by work you already did.
+7–8. Authors of the two most active Claude Code hook/plugin collections and awesome-lists (find via github.com/topics/claude-code). Template C.
+9–10. Two authors of recent posts/threads complaining about AI PR slop — search "AI generated pull requests" on HN/lobste.rs from the last month; the authors of the top two write-ups. Template A, referencing their specific post.
+
+Ship 2 per evening for a week. Log every send + result in [`OUTREACH_LOG.md`](../OUTREACH_LOG.md) — that log becomes your next Show HN ("What happened when I sent signed proof to 10 maintainers").
+
+---
+
+## Part 3 — Permissionless channels
+
+- **GitHub Actions Marketplace**: publish receipt-gate. A form, not a lottery. Requires: repo public ✅, v1 tag ✅, branding block in action.yml ✅. This is the single highest-leverage hour on this list.
+- **Repo topics**: added to both repos (ai-agents, ci, attestation, supply-chain, claude-code, verification, github-actions, sbom, ed25519, runtime-verification).
+- **Awesome-list PRs**: awesome-actions, awesome-claude-code, awesome-ai-agents. A PR is a contribution, not a promotion; new accounts are not penalized.
+- **dev.to / Hashnode write-up**: "My AI agent now hands me signed receipts" — the launch-kit Show HN body, expanded, GIF embedded. New authors are not filtered; posts rank in Google for months.
+- **X/Bluesky without followers**: don't broadcast — quote-reply into active AI-code-slop threads with the tamper-demo GIF. Replies ride the host thread's audience; follower count irrelevant.
+- **lobste.rs**: invite-only; one of your Part 2 replies is likely to be a user. Asking a warm contact for an invite is normal there.
+
+---
+
+## Part 4 — The math
+
+A Show HN is one lottery ticket: high variance, one draw, bouncer at the door. Ten receipt-first emails are ten independent draws at a far higher hit rate with zero gatekeeping, and every reply compounds: a user, a quote for the README, a lobste.rs invite, an issue filed by a stranger. The front page is not the start of distribution; it's the victory lap after receipts already exist in the wild.
+
+Rule for the log: a non-reply is `absence_of_signal`, not `rejected`. Fail closed on claims, never on morale.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bootproof",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "The honest run button for repos. Boots unfamiliar code when it safely can, tells the truth when it cannot, and signs proof of what happened.",
   "license": "Apache-2.0",
   "type": "module",
@@ -28,12 +28,18 @@
     "url": "git+https://github.com/bootproof/bootproof.git"
   },
   "keywords": [
-    "devtools",
-    "onboarding",
-    "docker",
-    "run",
     "attestation",
-    "reproducibility"
+    "runtime-verification",
+    "ai-agents",
+    "claude-code",
+    "ci",
+    "signed",
+    "receipt",
+    "proof",
+    "health-check",
+    "supply-chain",
+    "provenance",
+    "verification"
   ],
   "engines": {
     "node": ">=20.11"