public-release-verification
Verify whether a software release is publicly proven across repository identity, baseline, build, package, deployment, HTTP routes, SEO metadata, browser desktop/mobile behavior, locale content, visual availability, and human approval. Use when an agent is preparing or reviewing
Install
npx skills add https://github.com/markoblogo/abvx-agent-skills/tree/main/skills/public-release-verification
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install markoblogo-abvx-agent-skills@llmmart
git clone https://github.com/markoblogo/abvx-agent-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole markoblogo/abvx-agent-skills collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Public Release Verification
Use this skill to decide what a release has actually proved. A green build is one input, not a public-release claim.
Verification Contract
Capture one evidence row for each applicable domain:
- repository/commit identity — repository, branch/tag, commit, and clean or intentionally dirty baseline;
- baseline and diff — starting state, changed scope, and unrelated-change check;
- build/package proof — tests, build, package contents, and metadata;
- deployment proof — deployment provider, deployment identifier, and completed status;
- HTTP/route proof — actual public URL, status code, redirects, and key routes;
- SEO/metadata proof — title, description, canonical, robots, and social metadata where applicable;
- browser desktop/mobile proof — real browser observations, responsive states, interaction, console, and network failures;
- locale/content proof — requested locales, visible strings, fallback behavior, and content constraints;
- device/runtime proof — real-device or runtime-specific observations where the release target is hardware or a constrained runtime;
- visual/public availability proof — screenshots or equivalent observed public rendering and availability window;
- human approval and cutover status — named approval evidence, approval scope, and whether DNS, traffic, or publication gates remain open.
Mark each row PROVEN, PARTIAL, BLOCKED, or UNKNOWN, with command,
URL, artifact path, timestamp, and verifier where available. Use N/A only
when the row genuinely does not apply and explain why.
Claim Rules
- Build proof is not live proof.
- Deployment proof is not HTTP proof.
- HTTP proof is not browser or visual proof.
- Browser proof is not owner approval or public cutover authorization.
- Missing evidence is
UNKNOWN, not success. - A final
PROVENclaim requires every applicable release-critical row to bePROVENand the human gate to be explicit. - Report
PARTIALwhen useful evidence exists but a release-critical row is incomplete; reportBLOCKEDwhen a required check failed or access prevents verification.
Safe Route
- Record the repository identity and baseline before running release checks.
- Run the narrowest package/build checks, then collect deployment evidence.
- Verify the actual public URL and routes with the intended locale and viewport matrix.
- Inspect metadata, visible copy, console/network errors, and screenshots.
- Attach the evidence matrix and list unknowns, blockers, and human gates.
- Ask the owner for approval only after the evidence packet is reviewable.
Do not auto-publish, switch DNS, promote traffic, accept visual approval, or claim public availability on behalf of a human. Do not use a deployment log as a substitute for an observed public route.
Output
Release: <name/version>
Repository: <repo>@<commit/tag>
Status: PROVEN | PARTIAL | BLOCKED | UNKNOWN
<domain>: <status> — <evidence reference or reason>
...
Human gate: <approved / pending / not applicable, with scope>
Open blockers: <none or list>
Claim boundary: <what this packet proves and does not prove>
Composition
- Use
delivery-preflight-gatebefore a risky release run. - Use
browser-verificationfor the browser evidence row. - Use
confidence-fragility-reviewwhen the release summary sounds stronger than its evidence. - Use
skill-health-auditwhen reviewing this contract or its fixtures.
Files (abvx-agent-skills)
-
agents
-
openai.yaml 291 B
Loading…
-
-
SKILL.md 4 KB
--- name: public-release-verification description: Verify whether a software release is publicly proven across repository identity, baseline, build, package, deployment, HTTP routes, SEO metadata, browser desktop/mobile behavior, locale content, visual availability, and human approval. Use when an agent is preparing or reviewing a release and must separate build proof, live proof, and owner-gated public claims. license: MIT metadata: abvx_status: experimental abvx_origin: original abvx_eval_tier: rollout_checked --- # Public Release Verification Use this skill to decide what a release has actually proved. A green build is one input, not a public-release claim. ## Verification Contract Capture one evidence row for each applicable domain: 1. **repository/commit identity** — repository, branch/tag, commit, and clean or intentionally dirty baseline; 2. **baseline and diff** — starting state, changed scope, and unrelated-change check; 3. **build/package proof** — tests, build, package contents, and metadata; 4. **deployment proof** — deployment provider, deployment identifier, and completed status; 5. **HTTP/route proof** — actual public URL, status code, redirects, and key routes; 6. **SEO/metadata proof** — title, description, canonical, robots, and social metadata where applicable; 7. **browser desktop/mobile proof** — real browser observations, responsive states, interaction, console, and network failures; 8. **locale/content proof** — requested locales, visible strings, fallback behavior, and content constraints; 9. **device/runtime proof** — real-device or runtime-specific observations where the release target is hardware or a constrained runtime; 10. **visual/public availability proof** — screenshots or equivalent observed public rendering and availability window; 11. **human approval and cutover status** — named approval evidence, approval scope, and whether DNS, traffic, or publication gates remain open. Mark each row `PROVEN`, `PARTIAL`, `BLOCKED`, or `UNKNOWN`, with command, URL, artifact path, timestamp, and verifier where available. Use `N/A` only when the row genuinely does not apply and explain why. ## Claim Rules - Build proof is not live proof. - Deployment proof is not HTTP proof. - HTTP proof is not browser or visual proof. - Browser proof is not owner approval or public cutover authorization. - Missing evidence is `UNKNOWN`, not success. - A final `PROVEN` claim requires every applicable release-critical row to be `PROVEN` and the human gate to be explicit. - Report `PARTIAL` when useful evidence exists but a release-critical row is incomplete; report `BLOCKED` when a required check failed or access prevents verification. ## Safe Route 1. Record the repository identity and baseline before running release checks. 2. Run the narrowest package/build checks, then collect deployment evidence. 3. Verify the actual public URL and routes with the intended locale and viewport matrix. 4. Inspect metadata, visible copy, console/network errors, and screenshots. 5. Attach the evidence matrix and list unknowns, blockers, and human gates. 6. Ask the owner for approval only after the evidence packet is reviewable. Do not auto-publish, switch DNS, promote traffic, accept visual approval, or claim public availability on behalf of a human. Do not use a deployment log as a substitute for an observed public route. ## Output ```text Release: <name/version> Repository: <repo>@<commit/tag> Status: PROVEN | PARTIAL | BLOCKED | UNKNOWN <domain>: <status> — <evidence reference or reason> ... Human gate: <approved / pending / not applicable, with scope> Open blockers: <none or list> Claim boundary: <what this packet proves and does not prove> ``` ## Composition - Use `delivery-preflight-gate` before a risky release run. - Use `browser-verification` for the browser evidence row. - Use `confidence-fragility-review` when the release summary sounds stronger than its evidence. - Use `skill-health-audit` when reviewing this contract or its fixtures. -
SKILL_CARD.md 2.6 KB
Loading…
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.