V0.4.1/tweakking git story teller

[gimlichael]

This pull request introduces a minor release (v0.4.1) focused on significant enhancements to the git-story-teller skill, with improvements to grounding rules, context navigation, subagent delegation, evidence-based language, and deterministic output artifacts. The documentation has been updated to reflect these new features, stricter contracts, and expanded sync requirements for skill management.

Key changes:

git-story-teller Skill Enhancements

• Enforces complete-read grounding: Agents must fully read target and overview contexts before writing, using chunked files and indexes to handle output caps, and cannot skip evidence due to prompt/token limits. [1] [2] [3]
• Adds public API summaries and engineering signal maps to context output, helping agents focus on consumer-facing types, validation guards, and test evidence in their stories. [1] [2] [3]
• Supports subagent-friendly delegation: When multiple independent target contexts exist, subagents can process them in isolation, reducing prompt contention while maintaining strict grounding. [1] [2]
• Implements conservative test ownership mapping: Only assigns a test project if there's a dedicated, unambiguous match, avoiding accidental assignment of downstream tests. [1] [2]
• Introduces chunked context navigation: Emits *.context.index.md and ordered chunk files to enable robust, complete reading even when tools cap single-file output. [1] [2] [3]

Evidence-based Language and Grounding

• Forbids unmeasured behavior claims (e.g., "most common mistake") unless supported by explicit evidence in the context; mandates conditional language when evidence is lacking. [1] [2]
• Clarifies that context indexes are navigation aids only and cannot be used as evidence for story claims.

Documentation and Workflow Updates

• Updates SKILL.md, README.md, and CHANGELOG.md to document new output artifacts, workflow contracts, public API orientation, engineering signals, chunked navigation, and stricter grounding requirements. [1] [2] [3] [4] [5] [6]
• Expands the AGENTS.md skill sync section to include Gemini Antigravity install location, requiring sync across four locations for repo-managed skills. [1] [2]

Release Management

• Bumps version to 0.4.1 and adds a detailed changelog entry summarizing all new features, changes, and fixes. [1] [2]

---

These updates make git-story-teller more robust, deterministic, and evidence-driven, while improving agent workflow and output reliability.

[qodo-code-review]

Review Summary by Qodo

Strengthen git-story-teller with complete-read grounding and chunked context navigation

✨ Enhancement 🧪 Tests

Walkthroughs

Description

• Implements complete-read grounding with chunked context navigation
  - Emits *.context.index.md and ordered *.context.chunks/*.md files for robust reading when tools
  cap output
  - Enforces mandatory full-context reads before writing stories
• Adds public API summaries and engineering signal maps to context
  - Generates deterministic navigation aids highlighting consumer-facing types, lifecycle contracts,
  validation guards, and test evidence
• Implements conservative test ownership mapping
  - Prefers dedicated test projects with matching names over downstream package tests
• Enforces evidence-based language validation
  - Prevents unmeasured frequency claims like "most common" without explicit source evidence
  - Requires conditional phrasing for tradeoff guidance
• Supports subagent-friendly delegation for independent target contexts
• Updates skill sync guidance to include Gemini Antigravity install location

Diagram

flowchart LR
  A["Context Generation"] --> B["Full Context File"]
  A --> C["Context Index"]
  A --> D["Ordered Chunks"]
  B --> E["Agent Reading"]
  C --> E
  D --> E
  E --> F["Public API Summary"]
  E --> G["Engineering Signals"]
  F --> H["Target Story"]
  G --> H
  H --> I["Completed Target Results"]
  I --> J["Overview Synthesis"]
  J --> K["Result/Index.md"]

Loading

File Changes

1. skills/git-story-teller/scripts/story.cs ✨ Enhancement +649/-26

Implement chunked context navigation and API/signal generation

• Adds MaxContextChunkBodyBytes constant (36 KB) for chunking large contexts
• Implements WriteContextArtifactsAsync to generate full context, index, and ordered chunk files
• Adds SplitContextIntoChunks and SplitOversizedLine for deterministic context splitting
• Implements BuildContextIndex and BuildChunkFile to create navigation aids
• Adds DiscoverPublicApiTypes and BuildPublicApiSummary to generate API navigation tables
• Implements BuildEngineeringSignals with regex-based signal extraction for exceptions, lifecycle,
 hosting, and tests
• Refactors FindTestDirectory to use conservative test ownership mapping with TestProjectMatch
 record
• Adds IsOwnTestProjectName and StripKnownTestSuffix for dedicated test project matching
• Implements FilterLowSignalPackedContent to remove GlobalSuppressions.cs from packed context
• Updates manifest generation to include contextIndex and contextChunks entries
• Enhances instructions with complete-read contract, chunk navigation guidance, and evidence-based
 language rules
• Adds regex patterns PublicTypeRegex and PublicMemberRegex for API discovery

skills/git-story-teller/scripts/story.cs

---

2. AGENTS.md 📝 Documentation +7/-5

Add Gemini Antigravity to four-way skill sync guidance

• Expands skill sync guidance from three locations to four
• Adds ~/.gemini/antigravity/skills/<name>/ as required sync location
• Updates copy commands to include Gemini Antigravity path
• Clarifies that all four locations must stay synchronized

AGENTS.md

---

3. CHANGELOG.md 📝 Documentation +28/-1

Document v0.4.1 release with new features

• Adds v0.4.1 release entry with comprehensive feature list
• Documents complete-read grounding, subagent delegation, evidence-based language, public API
 summaries, engineering signals, conservative test mapping, and chunked navigation
• Lists all added features, changed behaviors, and fixes

CHANGELOG.md

---

View more (3)

4. README.md 📝 Documentation +12/-4

Update README with complete-read and signal generation features

• Updates git-story-teller description to mention public API summaries, engineering signals,
 conservative test paths, context indexes, and chunk files
• Expands "Why git-story-teller?" section with new capabilities
• Adds bullet points for public API first, engineering signal maps, conservative test ownership,
 low-signal filtering, chunked navigation, complete-read grounding, subagent-friendly targets,
 story-sourced overview, and phase-scoped reading
• Clarifies grounded prose rules to forbid unmeasured frequency claims

README.md

---

5. skills/git-story-teller/SKILL.md 📝 Documentation +86/-13

Refine skill contract for complete-read grounding and subagents

• Updates description to mention public API summaries, engineering signals, and chunked context
 indexes
• Adds critical rules for complete-read grounding and subagent delegation
• Clarifies that context indexes are navigation aids, not evidence
• Adds guidance on using generated API summaries and engineering signals as inspection prompts
• Expands workflow steps with complete-read requirements and chunk navigation
• Adds optional subagent strategy section for independent target contexts
• Enhances overview step with requirement to read completed target result files as primary source
• Adds grounding rules against unmeasured frequency claims
• Expands verification checklist with context index/chunk validation and frequency-claim detection
• Adds complete-read rules section replacing context budget rules
• Adds good output characteristics for subagent coordination and complete reads
• Expands anti-patterns with subagent misuse, capped context handling, and frequency-claim mistakes

skills/git-story-teller/SKILL.md

---

6. skills/git-story-teller/evals/evals.json 🧪 Tests +74/-5

Add evals for complete-read, subagents, and evidence-based language

• Updates eval 1 to require context index and chunk usage when full context is capped
• Updates eval 2 to recognize contextIndex and contextChunks manifest entries
• Updates eval 3 to require reading completed target result files as primary overview source
• Adds eval 7 testing complete-read grounding with capped file output and chunk navigation
• Adds eval 8 testing subagent-per-target delegation with independent contexts
• Adds eval 9 testing evidence-based tradeoff guidance without frequency claims
• Adds eval 10 testing public API summaries, engineering signals, and internal-detail usage
• Adds eval 11 testing conservative test ownership mapping with downstream test projects

skills/git-story-teller/evals/evals.json

---

[qodo-code-review]

Code Review by Qodo

Looking for bugs?

Check back in a few minutes. An AI review agent is analyzing this pull request.

[greptile-apps]

Greptile Summary

This PR delivers v0.4.1 of the git-story-teller skill, adding chunked context navigation (*.context.index.md + ordered chunk files), public API summary and engineering signal map generation, conservative test project ownership matching, low-signal file filtering (GlobalSuppressions.cs), subagent delegation patterns, and stricter evidence-based grounding rules. Documentation across AGENTS.md, README.md, CHANGELOG.md, and SKILL.md is updated accordingly, and the story.cs runner gains the corresponding C# implementation for all new artifacts.

The previous review's compiled-regex concern is fully resolved — PublicTypeExpression and PublicMemberExpression are now static readonly fields.

Confidence Score: 5/5

Safe to merge; only a single P2 eval-consistency finding with no runtime impact.

No P0 or P1 issues found. The C# script additions are well-structured, the compiled-regex regression from the previous review is fixed with static readonly fields, and the chunking/manifest logic is internally consistent. The sole finding is a contradictory first expectation in eval 11 that won't affect runtime behaviour.

skills/git-story-teller/evals/evals.json — eval 11 first expectation contradicts its own prompt scenario.

Important Files Changed

Filename	Overview
skills/git-story-teller/scripts/story.cs	Large addition of public API summary generation, engineering signal map, context chunking/indexing, conservative test ownership logic, low-signal file filtering, and updated manifest serialization; static compiled Regex fields replace per-call construction (addresses previous review comment); logic is sound.
skills/git-story-teller/SKILL.md	Substantial expansion of grounding rules, subagent delegation strategy, evidence-based language requirements, and chunked context navigation contract; well-structured and internally consistent.
skills/git-story-teller/evals/evals.json	Adds evals 7-11 and updates evals 1-3; eval 11 has a contradictory first expectation that says the dedicated test project exists when the prompt says it does not.
AGENTS.md	Adds Gemini Antigravity as a 4th sync location throughout; corrects heading from "both" to "local installs"; fully addresses previous P1/P2 review comments.
CHANGELOG.md	Adds 0.4.1 changelog entry dated 2026-05-04 (one day in the future relative to current date 2026-05-03); version links updated correctly.
README.md	Documentation updated to cover Gemini Antigravity skill folder and expanded git-story-teller feature bullet list; no issues found.

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
skills/git-story-teller/evals/evals.json:488-497
**Eval 11 prompt and first expectation contradict each other**

The `prompt` field explicitly states "there is no dedicated `Codebelt.Extensions.Xunit.Tests` project", but the first expectation says "Recognizes `Codebelt.Extensions.Xunit.Tests` as the dedicated test project for `Codebelt.Extensions.Xunit`". An agent running this eval in a grader will see conflicting signals: the scenario says the project doesn't exist, yet the expectation rewards recognising it as the dedicated test project. The remaining expectations (leave `Test path` undiscovered, don't use `Hosting.AspNetCore.Tests`) are consistent with the prompt — only the first expectation is out of place. Consider revising it to something like "Recognises `<PackageName>.Tests` as the expected dedicated-test-project naming convention" so it tests naming-pattern knowledge without implying the file exists in the scenario.

_{Reviews (2): Last reviewed commit: "⚡ Optimize regex compilation in story.cs" | Re-trigger Greptile}