Skip to content

Expose complete bounded OCR review tool usage in GitLab summary #142

Description

@xeonvs

Problem

The GitLab technical summary reports aggregate token usage and OCR tool-call activity, but the existing inline tool breakdown shows only the six most frequent entries. Less frequent review tools are collapsed into +N more, so an operator or manager cannot see whether OCR primarily read files, searched the repository, checked cross-file diffs, consulted toolkit context, or emitted review output.

Tool-call counts explain review activity only. OCR 1.10.0 does not provide per-tool token usage, so the toolkit must not claim that calls allocate the reported input/output/cache tokens.

Scope

Keep the existing one-line format:

- all OCR tool calls: N total (`tool`: count, ...)

Publish every admitted non-zero counter from this closed useful set:

  • OCR inspection: file_read, file_read_diff, file_find, code_search;
  • OCR output/lifecycle: code_comment, task_done;
  • toolkit context/evidence: ocr_toolkit_evidence, context_list, context_get.

Keep dynamic external MCP tools in the existing verified per-server summary. Do not publish their raw names.

Acceptance criteria

  • The existing inline wording and layout are unchanged.
  • The tool-call line is absent when the total is zero or the admitted breakdown is empty.
  • Every admitted counter greater than zero is shown, ordered by descending count and then tool name; more than six useful tools are no longer collapsed.
  • Zero, negative, boolean, malformed, oversized, unknown, and dynamic external-tool counters do not gain a public tool-name projection.
  • Aggregate token usage remains a separate line; documentation states that tool counts describe activity and are not per-tool token attribution.
  • Raw arguments, results, errors, paths, IDs, provider data, and dynamic external MCP tool names remain private.
  • DLP, receipt v5, findings, severity, posting transactions, telemetry ownership, and automatic approval are unchanged.
  • Focused formatter, posting, DLP, Markdown/control-character, bounds, and approval regressions pass.

Delivery boundary

Track in Draft PR #141 under milestone v0.8.3. Keep the PR Draft and leave this issue and the milestone open until the separately authorized stable release lifecycle completes.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions