feat(plugin): DTPR authoring studio — five skills + research corpus

[pichot]

Expands plugin/dtpr/ from a two-skill pair into a five-skill authoring studio (dtpr-describe-system, dtpr-datachain-structure, dtpr-category-audit, dtpr-element-design, dtpr-comprehension-audit) backed by a shared comprehension rubric, a file-based research corpus with append-only INDEX.md, and an extended verify.mjs that validates corpus frontmatter, INDEX append-only integrity against the origin/main merge base, and cross-sibling:* eval symmetry. dtpr-describe-system now accepts 0+ artifacts (PDFs, URLs, verbal) via a new Phase 0 trial-call-and-degrade tool probe with multi-host degradation for Claude Code / Cowork / Claude.ai. dtpr-schema-brainstorm is hard-deleted; its eval prompts port to the three schema-tier replacements and straddlers land as cross-sibling:* negatives so the router has explicit counter-examples for every peer. Plugin version bumps to 0.2.0 with a new CHANGELOG.md and a README rewrite that adds the per-host capability matrix, the References section (rubric + block template), and the Research corpus contract. Implements docs/plans/2026-04-20-001-feat-dtpr-authoring-studio-plan.md end-to-end; 77 eval prompts across 5 files, pnpm test:plugin passes.

[greptile-apps]

Greptile Summary

Expands the plugin/dtpr/ from a two-skill pair into a five-skill authoring studio (dtpr-describe-system, dtpr-datachain-structure, dtpr-category-audit, dtpr-element-design, dtpr-comprehension-audit), hard-deleting dtpr-schema-brainstorm and porting its eval prompts as cross-sibling:* negatives. The release adds a file-based research corpus with an append-only INDEX.md, a shared comprehension rubric, and an extended verify.mjs that validates corpus frontmatter, INDEX append-only integrity against the origin/main merge base, and cross-sibling eval symmetry across all five skills.

Confidence Score: 5/5

Safe to merge — all remaining findings are P2 style/completeness suggestions that do not affect runtime correctness.

Cross-sibling eval symmetry is fully verified programmatically and manually confirms IDs match across all five skills. The corpus verifier is well-structured with graceful degradation on missing git history. The two findings are an unused import and a missing reverse-index check that is a best-practice gap rather than a runtime defect.

plugin/dtpr/evals/verify.mjs — unused basename import and missing corpus-file→INDEX reverse check.

Important Files Changed

Filename	Overview
plugin/dtpr/evals/verify.mjs	Major extension: adds corpus-entry frontmatter validation, INDEX.md append-only integrity, and cross-sibling eval-symmetry checks. Two minor issues: unused basename import and missing reverse check (every corpus file must appear in INDEX.md).
plugin/dtpr/skills/dtpr-describe-system/SKILL.md	Significant expansion with Phase 0 trial-call-and-degrade, multi-artifact budgeting (inline/chunk-relevant/reject bands), multi-host degradation matrix, and research corpus integration. Sibling routing table and security framing are well-specified.
plugin/dtpr/evals/describe-system.evals.json	Adds five new positive prompts (PDF/URL/verbal artifact scenarios) and four cross-sibling negatives; cross-sibling IDs match their counterpart should_trigger entries in the sibling eval files.
plugin/dtpr/evals/element-design.evals.json	New file; 6 should_trigger, 8 should_not_trigger (4 cross-sibling). Provides the target positive IDs referenced by all four sibling eval files' cross-sibling entries.
.github/workflows/plugin-test.yaml	Adds fetch-depth: 0 to actions/checkout so git merge-base origin/main HEAD in verify.mjs has the full history it needs for the append-only check.
plugin/dtpr/research/README.md	New file; fully specifies the corpus contract (retrieval, write path, concurrency, append-only invariant, privacy prefix convention, authority tier enum).
plugin/dtpr/evals/schema-brainstorm.evals.json	Deleted; its should_trigger prompts were redistributed as cross-sibling negatives across the four replacement skills.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    User([User prompt]) --> Router{Skill router}

    Router -->|"describe / disclose / datachain"| DS[dtpr-describe-system]
    Router -->|"datachain-type shape / category add-retire"| DC[dtpr-datachain-structure]
    Router -->|"audit elements inside a category"| CA[dtpr-category-audit]
    Router -->|"draft / retire element"| ED[dtpr-element-design]
    Router -->|"grade for public comprehension"| CX[dtpr-comprehension-audit]

    DS -->|Phase 0| Probe[Trial-call-and-degrade\nRead · WebFetch · Task · Write]
    DS -->|Phase 2| MCP[(DTPR MCP\napi.dtpr.io)]
    DS -->|Research ctx| Corpus[(research/\nINDEX.md)]
    DS -->|Schema gap| ED
    DS -->|Category off| CA
    DS -->|Shape off| DC
    DS -->|Comprehension| CX

    subgraph verify.mjs
        V1[Skill frontmatter]
        V2[Eval JSON structure]
        V3[MCP tool drift]
        V4[Corpus frontmatter]
        V5[INDEX append-only]
        V6[Cross-sibling symmetry]
    end

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: plugin/dtpr/evals/verify.mjs
Line: 38

Comment:
**Unused `basename` import**

`basename` is imported from `node:path` but is never referenced anywhere in the file. This was likely added in anticipation of `entry.name` vs `basename(entry.path)` usage that never landed.

```suggestion
import { dirname, join } from 'node:path'
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: plugin/dtpr/evals/verify.mjs
Line: 508-511

Comment:
**No reverse-index check: orphaned corpus files go undetected**

`verifyIndex` validates that every slug row in `INDEX.md` points to an existing file, but the inverse is not enforced — a corpus entry file created in `research/` without a corresponding `INDEX.md` row will pass the verifier silently. Because retrieval is routed entirely through `INDEX.md`, such files are invisible to the skills at runtime and effectively wasted. Consider adding a check that every file in `corpusEntries` is referenced in `INDEX.md`.

```js
// After verifyIndex(), add:
for (const entry of corpusEntries) {
  if (!indexedSlugs.has(entry.name)) {
    fail(`${entry.path}: corpus file has no corresponding row in INDEX.md.`)
  }
}
```
(Requires extracting the validated slug set out of `verifyIndex`.)

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "feat(plugin): retire dtpr-schema-brainst..." | Re-trigger Greptile}

[greptile-apps]

Unused basename import

basename is imported from node:path but is never referenced anywhere in the file. This was likely added in anticipation of entry.name vs basename(entry.path) usage that never landed.

Suggested change

	import { basename, dirname, join } from 'node:path'
	import { dirname, join } from 'node:path'

Prompt To Fix With AI

This is a comment left during a code review.
Path: plugin/dtpr/evals/verify.mjs
Line: 38

Comment:
**Unused `basename` import**

`basename` is imported from `node:path` but is never referenced anywhere in the file. This was likely added in anticipation of `entry.name` vs `basename(entry.path)` usage that never landed.

```suggestion
import { dirname, join } from 'node:path'
```

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

No reverse-index check: orphaned corpus files go undetected

verifyIndex validates that every slug row in INDEX.md points to an existing file, but the inverse is not enforced — a corpus entry file created in research/ without a corresponding INDEX.md row will pass the verifier silently. Because retrieval is routed entirely through INDEX.md, such files are invisible to the skills at runtime and effectively wasted. Consider adding a check that every file in corpusEntries is referenced in INDEX.md.

// After verifyIndex(), add:
for (const entry of corpusEntries) {
  if (!indexedSlugs.has(entry.name)) {
    fail(`${entry.path}: corpus file has no corresponding row in INDEX.md.`)
  }
}

(Requires extracting the validated slug set out of verifyIndex.)

Prompt To Fix With AI

This is a comment left during a code review.
Path: plugin/dtpr/evals/verify.mjs
Line: 508-511

Comment:
**No reverse-index check: orphaned corpus files go undetected**

`verifyIndex` validates that every slug row in `INDEX.md` points to an existing file, but the inverse is not enforced — a corpus entry file created in `research/` without a corresponding `INDEX.md` row will pass the verifier silently. Because retrieval is routed entirely through `INDEX.md`, such files are invisible to the skills at runtime and effectively wasted. Consider adding a check that every file in `corpusEntries` is referenced in `INDEX.md`.

```js
// After verifyIndex(), add:
for (const entry of corpusEntries) {
  if (!indexedSlugs.has(entry.name)) {
    fail(`${entry.path}: corpus file has no corresponding row in INDEX.md.`)
  }
}
```
(Requires extracting the validated slug set out of `verifyIndex`.)

How can I resolve this? If you propose a fix, please make it concise.

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
🔵 In progress View logs	dtpr-ai	6c39b69	Apr 20 2026, 02:13 PM