✨ Add per-prompt Rudder test reports

[Vivekyy]

Adds a temporary Markdown report at the end of each Rudder test-generation run, grouping human-readable test links by their captured prompts. Prompt groups include exact prompt text and a concise representation of prior agent context from the database. The context helper now exposes previous agent output, with runtime coverage for that behavioral field. Validation passed with npm run typecheck, npm test, and npm run build.

---

Greptile Summary

Adds a per-prompt Markdown report to Rudder test-generation runs.

• Exposes captured previous agent output in context helper prompt records.
• Groups final tagged tests by their exact captured prompts and prior context.
• Adds runtime coverage for retrieving previousAgentOutput from a captured transcript.

Confidence Score: 4/5

The PR should not merge until legacy database reads, early-run report finalization, and completion telemetry ordering are corrected.

The helper can terminate on pre-migration state, early workflow exits omit a report required for every run, and telemetry can mark a run complete before required report generation succeeds.

Files Needing Attention: skills/rudder/scripts/context.mjs and skills/rudder/SKILL.md

Important Files Changed

Filename	Overview
skills/rudder/SKILL.md	Adds the prompt-report workflow, but early exits bypass it and completion telemetry precedes the report it claims was reached.
skills/rudder/scripts/context.mjs	Exposes previousAgentOutput but can fail against an existing database that has not received the column migration.
test/skill-runtime.test.ts	Correctly validates retrieval of previousAgentOutput from a fresh, fully migrated transcript-backed state database.

Sequence Diagram

sequenceDiagram
    participant R as Rudder workflow
    participant C as context.mjs
    participant D as Prompt database
    participant T as Telemetry
    participant F as Temporary report
    R->>C: Refresh run context
    C->>D: Read promptText and previousAgentOutput
    D-->>C: Captured prompt records
    C-->>R: Context JSON
    R->>R: Match final tagged tests to prompts
    R->>T: Record completion
    R->>F: Create Markdown report
    F-->>R: Temporary report path
    R-->>R: Present report first in final response

Loading

_{Reviews (1): Last reviewed commit: "✨ Add per-prompt Rudder test reports" | Re-trigger Greptile}

Greptile also left 3 inline comments on this PR.

[github-actions]

📦 No plugin release on merge

The plugin package, tag, and GitHub Release already exist for 0.1.3.
Bump package.json to release a new plugin version.

[ghost]

Ready to review this PR? Stage has broken it down into 4 individual chapters for you:

	Title
1	Expose previous agent output in context helper
2	Verify agent output retrieval in runtime tests
3	Define prompt report generation requirements
4	Integrate reporting into the Rudder workflow

_{Chapters generated by Stage for commit c011c1a on Jul 30, 2026 5:57pm UTC.}

[greptile-apps]

Legacy database query failure

When an existing Rudder database predates the previous_agent_output migration and has not subsequently been migrated, the read-only context helper unconditionally selects that column, causing SQLite to terminate the helper without emitting the context JSON required to generate the report.

[greptile-apps]

Early exits bypass reporting

When a run stops before the generation loop because intent is missing, declined, or not captured, execution never reaches this sole report step, so the run ends without the report and temporary path required by the new every-run contract.

[greptile-apps]

Completion recorded before reporting

When final tests succeed but temporary report creation or presentation subsequently fails, step 12 has already recorded the run as completed even though that status is defined as reaching the normal report, leaving telemetry inconsistent with the actual outcome.