REVIEW.md

# Authoring review — 0.1.0

Build prepared 2026-09-18; review finalized 2026-09-19. This record distinguishes completed checks from remaining tests.

| Check | Current result | Scope |
| --- | --- | --- |
| Standard plugin validator | Passed | Manifest shape, assets, and supported metadata. |
| Standard skill validators | Both passed | Frontmatter and skill structure; not behavioral correctness. |
| Runtime boundary and archive | Passed | Exactly two skills; no MCP, app dependency, or authoring-material leakage. Manifest is at ZIP root; file membership, byte equality, and archive CRC were checked. |
| Casino reading response | Reviewed; cited content checked | A completed authoring-agent response gives contextual readings with source roles and a bounded application limit. Parent review opened all three cited sources and inspected relevant passages. |
| 2 Peter reading trial | Incomplete | Independent runs were interrupted by usage limits. No completed answer is available for evaluation. |
| Chrysostom attribution trial | Incomplete | Research activity survived, but no completed independent answer is available for evaluation. It is not counted as a pass. |
| Bounded attribution illustration | Prepared and source-checked by the authoring assistant | Two web witnesses were inspected for a short example. This is not an independent trial or an exhaustive attribution search. |
| Ordinary Chat/mobile activation | Not tested | Explicit file loading in an authoring trial cannot establish host discovery. |
| User learning/retention | Not tested | Requires the reader's experience; technical checks do not establish this. |

The two initial drafts contain 1,794 and 1,776 whitespace-delimited words. The 0.9.16 theology entry contains 8,525. These are word counts, not token budgets or guaranteed host limits. Instruction recovery remains required.

The casino trial's original complete execution transcript was not retained in the parent's context across the interruption. Its answer and agent-authored evidence index survived. Later parent inspection checks its source support; it does not retroactively prove the original execution. Relevant witnesses are New Advent's identified translation of Trullo, Canons 50 and 102; the OCA's Trullo chapter; and Fr Luke Veronis's parish sermon, Casinos, Lotteries and the Resurrection. Their distinct roles remain visible in the saved answer.

Both skill UI files explicitly allow implicit invocation. That setting permits selection; it does not establish that a host will select the skill for every relevant simple question. A local personal-marketplace entry was created. It is not evidence of installation or activation in ordinary Chat or on the phone.

The installed 0.9.16 manifest and three original skill entries were checked against the recorded baseline hashes and remained unchanged. The approved 1.0.0 MCP submission was not edited or published.

Prepared prompts in TESTING.md are not counted as executed cases. No ordinary Chat installation, automatic activation, exhaustive attribution search, or guarantee of reliability is claimed. These checks do not establish public behavioral reliability. Next evaluate ordinary-question activation and complete R1 and V1 in fresh conversations under recorded settings.

## Publication packaging — 2026-09-19

Renamed the private predecessor to Orthodox Source Study (`orthodox-source-study`). Version remains 0.1.0. The two skill bodies are unchanged except their release branding, and all four artwork files are byte-identical. The private predecessor and earlier plugins are preserved. Public listing text and distribution instructions were updated. Structural validators and archive checks were rerun; no additional behavioral pass or host-activation success is claimed. Earlier trial records above refer to the private predecessor.

## Source prominence revision — 0.1.1

Source entries now lead ordinary reading responses. Introductory orientation is at most one sentence when needed; relevance notes are normally one short sentence. Unrequested closing synthesis is omitted. Requested synthesis is normally one compact paragraph, with explicit detailed requests and material distinctions preserved. Authentication retains a brief bounded finding followed by prominent witnesses. No evidence, role, or access safeguards were removed. Packaging validation is separate from behavioral evaluation; new regression prompts have not been executed.

## Listing revision — 0.1.2

Public title changed to Eastern Orthodox Source Study. Existing identifier `orthodox-source-study` is retained for update continuity. One everyday starter question is retained; two detailed study and attribution-research prompts replace the generic examples. Descriptions welcome readers at any level while emphasizing research. Skill procedures are unchanged from 0.1.1; only release/title lines changed. No new behavioral validation is claimed.

## Portal length correction — 0.1.3

The supplied submission screenshot rejected the 0.1.2 subtitle and starter prompts: subtitle maximum 30 characters; each prompt maximum 128. Earlier local validators did not enforce those portal limits. Shortened the subtitle and the two research prompts, retained the everyday casino question, and added explicit length checks to the builder. Skill procedures are unchanged apart from release metadata. The screenshot also reports a pending skills scan; it does not establish a scan failure or approval.

## Compact answer revision — 0.1.4

A user-supplied Chat High response remained an extended synthesis. The pasted answer does not establish the exact skill revision delivered or applied. Replaced permissive presentation guidance with a compact reading-list default: ordinarily 2–4 entries and about 150 words, short relevance phrases, no introductory theological verdict or explanatory/concluding sections. Detailed requests and material evidence retain adequate scope. Verification keeps a one-sentence bounded finding before witnesses. Structural validation does not establish host delivery or behavioral compliance. New regression prompts have not been executed.

## Source breadth and standing integration — 0.1.5

The authoring change removes the numeric source-range and total-word target from both research skills. Brevity is controlled per entry instead. Added a third self-contained assess-orthodox-standing skill with the 0.9.16 full warrant hierarchy and all eight spelled-out category/meaning pairs. Source fields carry compact linked readings. All category-specific fields, contrary evidence, bounded residual findings, and withholding rules remain required. No routine status banner, internal-code display, or closing essay is restored.

The inherited proposition and warrant sections are checked for exact retention. The output section changes placement and presentation only: source-bearing fields become prominent compact entries, without a duplicate bibliography or essay. Source-finding and verification route actual standing questions to the new entry; ordinary questions remain compact lists. This is a package and instruction integration review, not evidence of Chat activation or flawless application. Fresh-host behavioral tests remain pending. Earlier version-specific limits in this historical review are superseded by 0.1.5.

## Concise standing presentation — 0.1.6

The user preferred 0.1.4's consistently brief style and supplied a long toll-house response. The exact instruction-delivery record for that response is unavailable here; its verbosity cannot be causally attributed to a particular skill read. Replaced 0.1.5's expanded block presentation with one spelled-out standing/proposition line and concise linked evidence entries. Category meanings and warrant hierarchy remain. Category-specific evidence remains substantive, without mandatory separate Meaning/checklist fields or repeated method notes. Ordinary topics do not trigger standing just because contested. Simple answers regain a roughly 150-word target; no numerical source ceiling returns. Structural checks and warrant-preservation comparison are separate from fresh-host behavioral testing, which remains pending.

## Follow-up format revision — 0.1.7

The user supplied a three-turn free-account thread about toll houses and gambling. The initial answers already added unsolicited classifications and synthesis; the third legitimately requested applications but repeated analysis across sections, qualifications, a table, and conclusions. Its standing sentence combined a general proposition and a different application claim. This is evidence about the pasted output, not verified instruction delivery or a demonstrated account/model cause. External citations in that thread were not freshly authenticated during this maintenance task.

Replaced overlapping reading-presentation/follow-up rules with one per-turn choice of answer shape. Comparisons use compact linked entries or one table, shared qualifications appear once, and routine offers of broader synthesis are removed. Standing presentation now explicitly keeps separate propositions under separate findings and preserves the requested scope. The inherited proposition and warrant hierarchy and all eight category meanings are unchanged. Verification follow-ups retain bounded findings and avoid repeating the entire investigation.

Release checks cover plugin/skill structure, portal metadata lengths, archive membership and byte integrity, and preservation of the standing warrants. These do not establish improved behavior in Chat. The exact supplied three-turn sequence and a transfer case are prepared in TESTING.md; fresh-host execution remains pending. No installation or publication was performed.

## Eight-source reading target — 0.1.8

The user requested an eight-source aim, permission to return fewer when necessary, and a closing question offering another search. Topical reading instructions now implement that target without filler, duplicate mirrors, a hard maximum, or a claim that more sources must exist. A focused follow-up need not restart research. Standing and authentication retain their evidence-driven witness needs; a topical reading component can use the target, but no source count establishes authenticity or reception. Brief source notes and the prohibition of routine synthesis remain.

Version metadata and listing text were updated. The standing warrant hierarchy and category meanings, icons, subtitle, and sample prompts are preserved. Package/skill validation and archive integrity checks are separate from behavioral evaluation. The 0.1.8 tests are prepared for a fresh Chat run; no installation or publication is performed by this release preparation.