CHANGELOG.md

# 0.9.16 — complete instruction delivery and recovery

Baseline: tagged 0.9.15. The user supplied a Chat report of unused line continuation and another of complete reader text truncated by its outer output, followed by reported chunk recovery.

- Expanded only the first essential check, keeping recovery guidance early in every entry.
- Required following offered continuation notices with their documented parameters.
- Specified verbatim contiguous chunks across separate bounded tool returns when full text exists inside code but its wrapper output is truncated; several prints in one call still share its cap.
- Required documented storage or rereading the same revision, tracking actual delivered ranges, recovering gaps, and reducing chunks still truncated.
- Distinguished code-held data and coverage metadata from instruction text actually received by the assistant.
- Clarified one complete delivery of matching bundled procedures, supported recovery before withholding, and honest limits when recovery is unavailable or stops making progress.
- Kept diagnostic instruction dumps out of ordinary answers and later recovery separate from pre-answer access.
- Preserved all seven canonical procedures byte for byte, all other essential checks, stable names, public title, artwork, and invocation metadata.

The release does not assert one platform-wide output cap, add unsupported reader parameters, shorten the substantive method, install personal skills, or change an MCP service. TESTING.md records authoring checks separately from ordinary-Chat validation still needed after upload.

# 0.9.15 — essential checks before detailed procedures

Baseline: tagged 0.9.14. The supplied normal-Chat access audit reported truncated entry reads and a failure to apply some visible rules.

- Put compact evidence, transmission, Scripture, confessional, role, standing, and output checks at the beginning of each entry.
- Required incomplete-read detection and recovery through exposed controls, with disclosure and appropriate withholding when required content remains unavailable.
- Placed each skill's principal detailed procedure first after its introduction.
- Preserved all seven complete canonical procedures byte for byte, together with stable identities, the plain title, artwork, metadata, and implicit invocation.
- Kept one canonical early-controls file and a checked generation process.
- Tested the evidence and instruction boundaries with deliberately limited 900-word excerpts. This is not a measurement of Chat's truncation threshold.

The early controls do not guarantee that instructions are delivered or obeyed. Complete files may still be truncated. The previous answer violated visible instructions as well as missing more detailed instructions; both facts remain part of the diagnosis.

# 0.9.14 — complete procedures in each entry file

Baseline: the recovered installed 0.9.13 snapshot, including the user's removal of the visible version suffix.

## Instruction access

- Embedded all seven complete procedures into each SKILL.md and replaced supporting-file links with internal section links.
- Preserved focused discovery descriptions and three stable skill identities.
- Maintained canonical authoring modules and generated complete entries deterministically.
- Removed obsolete runtime reference copies. Runtime scripts and maintenance sources are excluded from the plugin ZIP.
- Kept the title **Eastern Orthodox Theology** and advanced the manifest and instructions to 0.9.14.
- Required access diagnostics to report actual reads and errors without treating rejected identifiers as proof of missing files.

## Evidence refinements

- Kept exact target wording, actor roles, and materially different variants separate.
- Applied transcription-provenance limits explicitly to original-language text as well as translations.
- Required evidence for derivation or transmission beyond chronological order and thematic resemblance.
- Preserved historical hypotheses as attributed hypotheses in openings, headings, tables, summaries, and recommendations.
- Applied source-by-source citation requirements explicitly to repeated summary claims.

The confessional framework, complete doctrinal-standing warrants and fields, Scripture fidelity, comparative coverage, calendar checks, adversarial rules, and personal spiritual-case boundaries remain. TESTING.md distinguishes the observed normal-Chat failure, current checks, and testing still needed after upload.

# 0.9.13 — skills-only workflow revision

Baseline: the user's uploaded 0.9.12 archive. This release does not derive from the MCP mirror or 0.10.0 test bundle.

## Workflow and packaging

- Replaced three long entry documents with focused task instructions and explicit expected output.
- Packaged complete local supporting procedures with every skill, removing reliance on a companion being separately activated.
- Maintained shared procedures once and generated matching local copies; added checks for drift, missing references, mismatched versions, and nonportable paths.
- Narrowed discovery descriptions to actual tasks rather than isolated keywords.
- Preserved the stable plugin and three skill names, author metadata, supplied artwork, and implicit invocation.
- Added the visible version to the plugin title and generalized the example prompts while preserving each explicit skill name.
- Removed the inherited `composerIconDark` manifest property because the current supplied plugin validator rejects it. The original image asset is retained.

## Corrections

- Resolved conflicting instructions about honoring a user's output wrapper and later deleting that wrapper.
- Made one public header and Answer label explicit in the entry documents after forward tests missed them in shorter replies.
- Removed the stock inference that an unverified quotation is a later embellishment; that explanation now requires its own inspected evidence.
- Kept categorical attribution judgments distinct from a scoped failure to authenticate wording, including in conclusions and citation advice.
- Distinguished supplied text from independently authenticated editions, assistant translations from published wording, and catalog identifiers from textual locators.
- Prevented a supplied interpretation about one actor from silently resolving an ambiguous clause about another actor.
- Retained the complete doctrinal-standing hierarchy, exact proposition/quantifier controls, independent recognition of alternatives, positive-evidence requirement for unresolved reception, and nonnegative residual limits.
- Kept requested-source coverage and citations within each table cell and standing field. Required tables are repaired instead of silently replaced.
- Clarified calendar validation and the difference between observed website behavior, a valid appointment, and durable reception.
- Removed blanket source-map assumptions about critical-edition accessibility, reliable selectable text, and universal liturgical usage.

## Preserved constraints

The Orthodox confessional framework, mutually interpreting witnesses of Holy Tradition, honest reporting of adverse evidence, source-role distinctions, explicit adversarial activation, and personal spiritual-case boundaries remain. This release requires no MCP connection or persistent computer/server process.

## Validation boundary

Structural validators check packaging and instruction-file integrity. Independent forward tests exercise actual responses after direct loading of these files. Neither establishes automatic activation or complete compliance in normal Chat; the installed draft still needs testing there.