# Evidence index and image captions

Status date: 27 October 2026. The twelve numbered PNGs reproduce the final English slides. The native presentation retains editable text and workflow objects. Screenshots are actual process captures, not generated images or mock-ups.

| Slide | Source record | Caption and alternative text |
| --- | --- | --- |
| 01 | Current OpenAI Platform draft capture and connected manifest | PALO Governance development overview beside the real draft listing. The visible status is incomplete, needs attention and not published. |
| 02 | Six packaged SKILL.md files and proposal workflow descriptions | Six documentation workflows and their concrete outputs, from a lifecycle dossier to data provenance and fitness requirements. |
| 03 | Connected ZIP inventory, manifest and SHA-256 | Build diagram linking canonical references, autonomous skill directories, validation and a frozen draft archive. File tree and hash identify the package. |
| 04 | Actual palo-agentic-action-design/SKILL.md in the connected ZIP | Skill excerpts distinguish a future action from a past incident and separate authorization, execution, observed effect and business outcome. |
| 05 | Actual connected mcp.json, with private identifiers omitted | Abbreviated OAuth configuration beside a schematic of the intended ChatGPT-to-Reader request. This flow is not evidence of a completed authenticated session. |
| 06 | Microsoft Entra permission capture, 26 October | Cropped permissions table shows delegated palo:guide and palo:knowledge:read, both granted. Runtime client admission remains a separate unresolved step. |
| 07 | Current OpenAI Platform skills capture, 27 October | All six actual skill rows show Checks passed. This does not establish publication approval or runtime qualification. |
| 08 | host-tests/RESULTS.md and corrected harness evidence, 26 October | Static-package local tests: 16 scoped passes, one synthetic no-echo failure and one infrastructure-blocked case. Connected Reader tests remain separate. |
| 09 | reviews/round2/REVIEW-RESULT.md and retained reviewer transcripts | GLM and Gemini review scopes, accepted corrections and a false character claim rejected after checking the file bytes. |
| 10 | Current Platform MCP capture and owner-supplied Microsoft error | Real MCP status shows a verified domain and authentication still required. Diagram records the AADSTS500113 failure and missing exact callback. |
| 11 | Help Center escalation capture and read-only hosting/log checks, 27 October | Actual support acknowledgement beside hosting and logging observations. The escalation is not a technical answer. Private contact details are outside the crop. |
| 12 | Connected test plan and current submission record | Remaining qualification workflow: callback and client admission, eight connected cases, reviewer access, demo and public policies. |

## Provenance and scope

The static local test package is version 0.1.0. The connected draft is version 0.2.0. Results are not transferred between the two. The Agentic Action Design source retains a 0.1.0 reference snapshot label inside the later connected archive. The date describes packaging, not renewed verification of every external source.

The connected archive has 185 files and SHA-256 c453a94f736b2c38750023d646161083311e5b2fd64b27bffa2bdee1fe7440e3. Integrity checks identify bytes; they do not establish the truth of claims or successful runtime behavior.

The GLM review covered an extracted manifest, six skills and a shared contract. The successful Gemini review covered three critical skills. The reviews did not include the whole canonical corpus, authenticated tool calls or final connected archive approval. Their original records predate some later local test execution; the newer test report supplies the test outcomes shown here.

## Public references

- Project and author attribution: https://paloframework.org/
- Plugin submission process: https://developers.openai.com/plugins/deploy/submission
- Plugin authentication: https://developers.openai.com/plugins/build/auth
- Author and media contact: https://www.linkedin.com/in/fdegni/

The first three pages were accessible during preparation on 27 October 2026. The LinkedIn URL was supplied by the author. Platform and support screenshots document the author's own project state; they are not statements of endorsement by OpenAI or Microsoft.

## Publication boundary

Use the final slide PNGs and PDFs in this kit for distribution. Raw source screenshots and private support transcripts are intentionally excluded from the ZIP. The live editable presentation is an authoring copy and can retain source image bytes outside a visible crop. Exported slide images flatten the visible crop. Do not distribute the raw editable source without checking embedded media.