V0.4.2/git story telling rename#15
Conversation
- Context indexes now list real Markdown headings where available - Headingless packed-content chunks get inferred labels: Source Code, Test Coverage, NuGet Documentation, Project Metadata - Added eval 12 to validate deterministic label assignment per chunk content - Agents can navigate large evidence sets with stable, meaningful chunk descriptions
…xt chunks - Replace (none) placeholders with inferred labels based on packed file paths - BuildInferredChunkHeading() analyzes file composition to assign: Source Code, Test Coverage, NuGet Documentation, Project Metadata, Documentation, Repository Metadata - ClassifyPackedFilePath() maps files to categories: test/, src/, .nuget/, README.md, Directory.Build.*, .csproj - Enables agents to navigate large context chunks with meaningful, deterministic descriptions
- Added three new Changed bullets detailing context index chunk label inference - Documents replacement of (none) placeholders with deterministic labels - References new BuildInferredChunkHeading() and ClassifyPickedFilePath() implementation - Notes eval 12 coverage for chunk label assignment validation - Maintains 0.4.2 accumulated feature narrative with today's date (2026-05-05)
…-path labels - Updated skill table description to reference Contents column combining headings and labels - Refined SKILL.md documentation: context indexes list markdown headings plus packed-path labels in Contents - Updated eval 12: test Contents column combining both heading types, no ellipsis truncation - Clarified that chunks can include both real headings and inferred labels in same semicolon-delimited cell
… and packed-path labels - Renamed Headings column to Contents in context index table - Refactored BuildInferredChunkHeading() → BuildChunkContents() to combine both sources - Implemented ExtractInferredChunkLabels() generator yielding all path-derived categories - Removed ellipsis truncation (...) from Contents cell to show complete label list - Contents cell uses semicolon-delimited format for all entries: headings first, then labels
Greptile SummaryThis PR renames the Confidence Score: 4/5Safe to merge; changes are a well-structured rename with intentional convention narrowing and no logic regressions. No P0 or P1 findings. Two P2 observations: incomplete chunk label taxonomy in SKILL.md and a silently dropped plural Tests.cs heuristic in IsProbablyTestFile. Both are low-risk under Codebelt conventions. skills/git-repo-digest/SKILL.md (label documentation gap) and skills/git-repo-digest/scripts/digest.cs (IsProbablyTestFile narrowing) Important Files Changed
Flowchart%%{init: {'theme': 'neutral'}}%%
flowchart TD
A["digest.cs runner\n(dotnet run --file)"] --> B["ParseOptions\n--repo-url / --output-root"]
B --> C["CloneRepositoryAsync\ngit clone --depth 1"]
C --> D["DiscoverPackages\nsrc/ only"]
D --> E["FindTestDirectory\ntest/ only\n.Tests / .FunctionalTests"]
E --> F["BuildPackageContextAsync\nper package"]
F --> G["WriteContextArtifactsAsync\n*.context.md"]
G --> H["BuildContextIndex\n+ BuildChunkContents"]
H --> H1["ExtractHeadings"]
H --> H2["ExtractInferredChunkLabels\nClassifyPackedFilePath"]
H1 --> H3["Distinct + Join → Contents"]
H2 --> H3
H3 --> I["*.context.index.md\n*.context.chunks/*.md"]
F --> J["BuildOverviewContextAsync\nREADME + src/**/*.csproj\n+ test/**/*.csproj"]
J --> K["overview.context.md"]
G --> L["WriteManifestAsync\nmanifest.json\npackages + targets alias"]
K --> L
L --> M["Agent reads manifest\n→ writes result/{PackageName}.md\n→ writes result/Index.md"]
|
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
|
|
||
| For each target in the first phase: | ||
| For each package in the first phase: | ||
|
|
There was a problem hiding this comment.
The critical section lists Source Code, Test Coverage, NuGet Documentation, and Project Metadata as the stable packed-path labels, but ClassifyPackedFilePath in digest.cs can also emit Documentation (for root-level README.md files) and Repository Metadata (the catch-all for any path that doesn't match other patterns). Agents reading SKILL.md to understand the Contents column won't be told about these two additional labels. The eval contract (eval 12) does mention Documentation, but neither the eval nor SKILL.md documents Repository Metadata.
Suggest extending the "such as" list to include all six possible labels so the documentation matches what the index can actually contain.
Prompt To Fix With AI
This is a comment left during a code review.
Path: skills/git-repo-digest/SKILL.md
Line: 157
Comment:
**Chunk label taxonomy is incomplete**
The critical section lists `Source Code`, `Test Coverage`, `NuGet Documentation`, and `Project Metadata` as the stable packed-path labels, but `ClassifyPackedFilePath` in `digest.cs` can also emit `Documentation` (for root-level `README.md` files) and `Repository Metadata` (the catch-all for any path that doesn't match other patterns). Agents reading SKILL.md to understand the `Contents` column won't be told about these two additional labels. The eval contract (eval 12) does mention `Documentation`, but neither the eval nor SKILL.md documents `Repository Metadata`.
Suggest extending the "such as" list to include all six possible labels so the documentation matches what the index can actually contain.
How can I resolve this? If you propose a fix, please make it concise.2 participants
This pull request implements a comprehensive rename and terminology update for the
git-story-tellerskill, which is now calledgit-repo-digest. The changes affect documentation, skill descriptors, installation instructions, and the core narrative throughout the project. Additionally, the update clarifies and improves how context chunk labeling and test project discovery work, aligning with Codebelt conventions and making context navigation more meaningful and deterministic.Key changes include:
Skill and Terminology Rename:
git-story-tellertogit-repo-digestacross all references, including documentation, installation examples, file names (e.g., runner script renamed fromscripts/story.cstoscripts/digest.cs), and terminology withinSKILL.md,README.md, and the changelog. [1] [2] [3] [4] [5] [6]Documentation and README Updates:
Context Index and Chunk Labeling Improvements:
Test Project Discovery and Ownership:
src/, owned tests fromtest/, and specific suffix patterns, avoiding ambiguous or overly broad test assignments. Updated documentation to reflect this strategy. [1] [2] [3]Changelog and Evaluation Contracts:
These changes make the skill's purpose clearer, improve context navigation for agents, and align the project with Codebelt's deterministic and maintainable conventions.