v0.1.7 — Bounded internal PNG cache

This release bounds the internal natal PNG cache to prevent unbounded disk growth.

Changes

• Default internal cache budget: 512 MiB (CHART_CACHE_MAX_BYTES=536870912). Expired entries are removed first, then the oldest writes are evicted as needed.
• Default entry lifetime: 7 days from creation (CHART_CACHE_TTL_SECONDS=604800). Cache reads do not extend that lifetime. Maintenance runs on cache access/write; idle or stopped services do not run a cleanup timer.
• Atomic writes, cross-process coordination and a fixed set of locks keep concurrent cache operations bounded. Oversized images are returned without being cached.
• Public image URLs, calculation results and the 48-tool interface remain unchanged. This limit applies to internal cached PNG payloads, not public images, databases, logs or a filesystem-wide quota.

Validation

• GitHub Actions passed on Ubuntu and macOS for this exact source commit.
• All 25 cache regression tests passed in the production candidate environment.
• Before/after natal calculations and PNG bytes matched for Placidus and whole-sign examples; repeated requests used the cache.
• The maintainer-hosted HTTPS MCP was deployed and verified with initialization, tool discovery, repeated chart rendering, image download and an existing image URL.

Install and verify

Download hellenicstar-mcp-source.tar.gz, source-manifest.json and SHA256SUMS. Run sha256sum -c SHA256SUMS (or shasum -a 256 -c SHA256SUMS on macOS) before extracting, then follow docs/quickstart.md.

Source commit: b1b670cb740d1c388c5820b59f73d890a4e9a316.

This source release does not automatically upgrade existing local installations or third-party hosted instances. Glama releases are managed separately.

v0.1.6 — Clear conventions and smoother MCP calls

Firdaria requests now preserve the requested tradition name and explicitly disclose the effective implementation. The existing al_biruni alias maps to abu_mashar rules; this release does not add an independent historical algorithm or change period dates. Old calculation references whose source payload changed should be recalculated.

Tool schemas now describe known include fields and supported verification value shapes. Qualified city searches proceed when exactly one database name matches; ambiguous names and unverified regions still require clarification. Sky guidance distinguishes a moment/day snapshot from a complete annual retrograde calendar.

Validation: 1,313 local regression tests passed; 13 actual AI scenarios were exercised in 14 runs, retaining the initial annual-calendar failure and its follow-up. Professional direct calls, parameter choices and full output remain available. Large responses may still be truncated by clients. The Mini Program is outside this release.

See docs/validation/usability-fixes-2026-10-25.md for results and limitations.

v0.1.5 — Direct professional use and focused reading

Explicit professional requests can call tools directly without a required workflow-discovery step. Ordinary requests receive guidance for interpreting intent, preserving multiple techniques, and selecting relevant relationship fields. Explicit full-data requests retain full output; actual client truncation is distinguished from generic verification guidance.

All 48 tool names, schemas, defaults and calculation bodies remain unchanged. No Mini Program code changes.

Validation: 1,299 local tests passed; eight actual AI client scenarios covered four professional and four ordinary-language requests. These are scenario checks, not a cross-model reliability benchmark. Large full-data responses can still exceed client display limits.

v0.1.4 — Clearer workflow routing

Fix workflow selection for parallel chart requests, transit time ranges, data-only and image-only requests. Missing inputs remain explicit clarification steps; existing conversation information should be reused.

All 48 tools and input schemas are retained. No calculation engine, interpretation rules or licensing changes.

Validation: 68 targeted tests and 1299 full-suite tests passed locally, with one existing upstream deprecation warning. CI, deployment and platform discovery are tracked separately in docs/validation/workflow-routing-release-2026-10-25.md.

Linux and macOS CI passed. Production cutover and public HTTPS routing, calculation, clarification and PNG delivery checks passed. Glama 0.1.4 is published; latest observed TDQS is 3.9/5 (A), not a score improvement. Mini Program process and deployment were unchanged.

v0.1.3 — clearer tool inputs and task selection

Six tools now explain required birth/place inputs, timezone-aware moments, reference dates, output formats and which related tool to select. Added a bilingual task-selection table for sky, transit, lunation, time-lord and Jones workflows.

No tools removed or renamed; parameter types, defaults and calculation code are unchanged. Existing AGPL licensing and voluntary contribution guidance remain in place.

Validation: 1243 tests passed locally; all six affected tools completed real calls, including three PNG outputs, and clarified missing inputs. All 48 schemas are unchanged after excluding descriptions. CI and deployment evidence are tracked in docs/release-status.md.

0.1.2 — Tool guidance and community contributions

Clarifies how to call and choose five astrology tools: house rulers, solar-arc calculation and rendering, composite rendering, and solar-return rendering. Birth-time/location prerequisites, explicit house/residence choices, age precedence, result delivery and sibling-tool routing are documented without changing calculation behavior.

Includes bilingual contribution guidance, optional bug/feature/use-case templates and a PR template. AGPL-3.0 remains unchanged; feedback is voluntary and contributors retain copyright.

Validation: 1,243 local tests passed without skips. GitHub Actions 40150198569 passed on macOS and Ubuntu. All 48 schemas and annotations are unchanged after removing description keys; complete-file AST comparison confirms only targeted description metadata changed.

Glama 0.1.2 is now published (2026-10-25 10:09 Asia/Shanghai). Its average TDQS is 4.1/5 across 48 evaluated tools, up from 4.0 for 0.1.1. The five updated tools score 4.6, 4.7, 4.8, 4.8 and 4.8 respectively. These are tool-definition quality scores, not calculation-accuracy certification.

Production verification: the standalone HTTPS MCP service now runs this commit. A real client initialized, discovered all 48 matching tool schemas/descriptions, completed five calls, received three inline PNGs and downloaded all three image URLs successfully. New/old server calculation and rendering outputs were hash-identical. Mini Program process and start time were unchanged.

Source archive, per-file manifest and SHA256SUMS are attached.

v0.1.1 — Clearer MCP tool definitions

This source release improves MCP tool discovery and invocation guidance without changing calculation rules, input types, defaults, or validation constraints.

• Clarify every input of natal-priority calculation and Saturn/Jupiter-cycle rendering.
• Explain when to choose the graph, chart, data-table, or image tools.
• Preserve relationship_views discovery and all interpretation boundaries.
• Correct shared summary guidance: scalar membership lists stay complete, while row previews explicitly identify truncation.

Glama TDQS was 3.9/5 (A) for the previous platform release; this release does not claim a higher score before re-evaluation.

Validation: GitHub Actions run 40050682033 passed on Ubuntu and macOS, with 1,243 tests passing on each platform and one upstream deprecation warning.

The source archive includes an individual-file manifest and SHA256SUMS. It excludes credentials, local databases, runtime caches, and Git history.

Production verification (2026-10-24): the standalone HTTPS MCP endpoint now runs source cf288d1. All 48 tools retain their input types, defaults, constraints and annotations; updated descriptions are visible online. For default orb and orb_view=8, both full and summary preserve the same calculation reference and all 9 selected members. Original schema/member claims are supported and the deliberately wrong schema is contradicted in all four cases. The Mini Program process was not restarted. Glama 0.1.1 was also published; its new TDQS evaluation is pending.

v0.1.0 — verified source installation

首个 GitHub 源码发行版。包含当前 48 个 MCP 工具、本命重点 v4、关系观察 v0.14，以及 calculation-reference-v2 的跨 JSON 数字序列化修复。

This is the first GitHub source distribution. Runtime schemas retain their own versions; the GitHub tag and Glama release numbering do not replace them.

安装与分发 / Installation

• 中英文共用安装顺序与依赖说明；收紧自动安装、验证覆盖及精度表述。
• 新增 scripts/smoke_install.py：无需地名库，通过真实 stdio 完成坐标解析、本命计算、正确/错误字段核验和 PNG 检查。
• 源码包补齐英文 README、Glama 构建配置、维护者配置和快速开始。
• 附源码 tar.gz、逐文件 manifest 与 SHA256SUMS；不包含地名数据库、凭据、运行缓存或本地 Git 历史。

快速开始 / Quick start

Compatibility

• Current CI targets Python 3.12 on macOS and Linux. Native Windows is not covered by that matrix.
• New references use calculation-reference-v2; unchanged v1 references remain accepted. Recalculate if an old v1 reference has already been altered by JSON number serialization.
• Summary results retain complete scalar membership lists and explicitly mark truncated object-row lists.
• No calculation or interpretation policy changes were made in the distribution update. Claim verification checks structured fields, not psychological interpretation or independent algorithm correctness.
• Optional GeoNames data is built separately. Coordinate-based operation needs no city database or map API key.

Validation

• Source archive SHA-256 and each manifest entry verified after extraction; extracted-source stdio smoke passed.
• Fresh Python 3.12 environment with only requirements.lock passed the installation smoke.
• Four targeted packaging/install regression tests passed locally.
• macOS and Ubuntu CI each passed 1243 tests, zero skips, with one upstream deprecation warning.
• CI for source commit 91b721a
• Prior production calculation-repair evidence

Glama's initial 0.1.0 image uses commit 2a04efb, the preceding runtime-equivalent source. Its 48 tool definitions were evaluated at an average TDQS of 3.9/5 on 2026-10-24. TDQS measures tool definitions, not astrology algorithm accuracy. Smithery publication is not established by this release.