DOJ-3828: Migrate 12 agents from dojo-academy (PR-A of DOJ-3709)

[lapc506]

Summary

• Migrates 12 agent .md files from dojo-academy to IDT (PR-A of 2 in DOJ-3709 chain)
• Each agent migrates as-is per DOJ-3708 chain precedent (no decomposition; prose, frontmatter, Spanish accents preserved verbatim)
• After this PR merges, conceptual references like text-class-writer, content-reviewer, quiz-generator from the DOJ-3708 chain commands resolve to real agents in IDT
• dojo-academy source files NOT deleted (cleanup is DOJ-3711)

Files

12 new files in agents/:

• challenge-designer
• content-architect
• content-reviewer
• framework-extractor
• proofreader
• quiz-generator
• repo-analyzer
• research-agent
• student-perspective
• text-class-writer
• translation-reviewer
• translator

No name collisions with the 7 existing IDT agents (business-context-detector, changelog-generator, cmi5-metadata-writer, course-visualizer, learner-profile-builder, session-type-detector, slides-renderer). Total IDT agent count post-merge: 19.

Consumer-convention framing

Per the DOJ-3708 chain precedent, dojo-academy-specific framings are wrapped so the agents stay reusable across consumer plugins:

• content/courses/... paths → described as consumer-specific (dojo-academy convention shown)
• content/_templates/... → described as consumer-shipped templates
• nanobanana-output/ → described as consumer-specific holding folder
• skills/academy-philosophy/..., skills/blooms-taxonomy/..., skills/ship-first-design/... → described as consumer overlays (these skills migrate via DOJ-3829, PR-B of this chain)
• Claude Partner Network framing → scoped to dojo-academy as the example, not asserted as universal

Discrepancy with parent

Parent DOJ-3709 description title says "11 agents" but the table lists 12. PR-A migrates the 12 from the table (the source of truth).

CI verification (local)

All 3 DOJ-3774 lint checks pass on this branch's HEAD:

• check_frontmatter.py: 58 file(s) valid, 2 skipped (no frontmatter)
• check_agent_references.py: 19 agent(s) indexed, all IDT-internal references resolve
• check_json_schemas.py: 3 file(s) valid

Test plan

• DOJ-3774 CI workflow passes on PR (frontmatter + agent reference resolution + JSON schema)
• Greptile review hits 5/5 functional
• DOJ-3829 (PR-B skills) starts after this merges

Out of scope

• Skills migration — DOJ-3829 (PR-B)
• Cleanup in dojo-academy — DOJ-3711
• Convert academy-philosophy + content-standards to overlays — DOJ-3710

Closes DOJ-3828

Created by Claude Code on behalf of @andres

[lapc506]

@greptileai review

[greptile-apps]

Greptile Summary

This PR migrates 12 agent .md files from dojo-academy into the IDT agents/ directory as-is, bringing the total IDT agent count to 19 and resolving cross-references from the DOJ-3708 command chain. The migration is clean — no name collisions, frontmatter is valid across all 12 files, and consumer-specific paths/overlays are correctly framed as consumer conventions rather than IDT defaults.

Confidence Score: 4/5

Safe to merge — all 12 agents are additive, no existing files are modified, and CI lint checks pass.

The migration is faithful and well-scoped. The main findings are a '5 questions' / 6-row mismatch in the content-reviewer rubric and a hint template that only scaffolds 2 entries while the spec text references 'Hint 3.' Both are copy issues from the source that would cause minor confusion in practice. The research-agent's undeclared Context7 MCP dependency is worth documenting explicitly for downstream consumers.

agents/content-reviewer.md (Check 4 question count), agents/challenge-designer.md (hint template vs spec text), agents/research-agent.md (Context7 plugin dependency)

Important Files Changed

Filename	Overview
agents/content-reviewer.md	New agent for quality review; Check 4 header says "5 questions" but the rubric table contains 6, creating a confusing scoring discrepancy
agents/challenge-designer.md	New agent defining challenge/BUILD deliverable design; minor inconsistency: hint spec mentions 3 hints but template only scaffolds 2
agents/research-agent.md	New deep-dive research agent; uses Context7 MCP tools (plugin-specific) and WebSearch/Agent tools; Context7 dependency will fail if not installed in consumer environment
agents/content-architect.md	New agent for course/module structure design; references ${CLAUDE_PLUGIN_ROOT} commands not yet in IDT (out of scope per PR, consumer-dependent)
agents/text-class-writer.md	New primary content-writing agent; correctly labels Claude Partner Network and gemini-3-pro-image-preview as dojo-academy-specific; anti-pattern list is comprehensive
agents/translator.md	New Spanish LATAM translation agent; mandatory accent enforcement and glossary self-check steps are strong operational guards

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    RA[research-agent] --> FE[framework-extractor]
    RA --> CA[content-architect]
    RA --> TCW[text-class-writer]
    REPO[repo-analyzer] --> FE
    REPO --> TCW
    CA --> TCW
    CA --> QG[quiz-generator]
    CA --> CD[challenge-designer]
    FE --> TCW
    FE --> QG
    TCW --> CR[content-reviewer]
    QG --> CR
    CD --> CR
    CR --> PRF[proofreader]
    PRF --> TR[translator]
    TR --> TRV[translation-reviewer]
    SP[student-perspective] --> CR

Loading

Prompt To Fix All With AI

Fix the following 3 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 3
agents/content-reviewer.md:65
The header says "5 questions" but the rubric table contains 6 rows. An agent following these instructions will be confused about whether it has completed the evaluation after row 5. The `text-class-writer.md` peer agent (which uses the same rubric) correctly labels this as "six questions" — update the count here to match.

```suggestion
Score the text class on 6 questions (1-5 each):
```

### Issue 2 of 3
agents/challenge-designer.md:146-151
The spec text says "Hint 3 is more direct," implying up to 3 hints, but the template block only scaffolds 2 `

## Submission
```

### Issue 3 of 3
agents/research-agent.md:59-61
**Context7 MCP dependency not declared in frontmatter**

Step 2 instructs the agent to call `mcp__plugin_context7_context7__resolve-library-id` and `mcp__plugin_context7_context7__query-docs`, but these MCP tools are not listed in the frontmatter `tools:` field and depend on the consumer having the Context7 plugin installed. In an IDT consumer without Context7, this step will fail at runtime with no graceful fallback. Consider adding a conditional note ("if Context7 is installed, use …; otherwise fall back to WebFetch") to make the dependency explicit.

_{Reviews (1): Last reviewed commit: "DOJ-3828: Migrate 12 agents from dojo-ac..." | Re-trigger Greptile}

[greptile-apps]

The header says "5 questions" but the rubric table contains 6 rows. An agent following these instructions will be confused about whether it has completed the evaluation after row 5. The text-class-writer.md peer agent (which uses the same rubric) correctly labels this as "six questions" — update the count here to match.

Suggested change

	Score the text class on 5 questions (1-5 each):
	Score the text class on 6 questions (1-5 each):

Prompt To Fix With AI

This is a comment left during a code review.
Path: agents/content-reviewer.md
Line: 65

Comment:
The header says "5 questions" but the rubric table contains 6 rows. An agent following these instructions will be confused about whether it has completed the evaluation after row 5. The `text-class-writer.md` peer agent (which uses the same rubric) correctly labels this as "six questions" — update the count here to match.

```suggestion
Score the text class on 6 questions (1-5 each):
```

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

[greptile-apps]

The spec text says "Hint 3 is more direct," implying up to 3 hints, but the template block only scaffolds 2 <details> entries. An agent reading the template will consistently produce 2-hint challenges even when 3 would be appropriate. Either add a third placeholder or remove the "Hint 3" reference from the spec text.

Suggested change

	<details>
	<summary>Hint 2</summary>
	{Address a common stumbling point}
	</details>
	## Submission
	<details>
	<summary>Hint 2</summary>
	{Address a common stumbling point}
	</details>
	<details>
	<summary>Hint 3</summary>
	{More direct nudge toward the solution path}
	</details>
	## Submission

Prompt To Fix With AI

This is a comment left during a code review.
Path: agents/challenge-designer.md
Line: 146-151

Comment:
The spec text says "Hint 3 is more direct," implying up to 3 hints, but the template block only scaffolds 2 `<details>` entries. An agent reading the template will consistently produce 2-hint challenges even when 3 would be appropriate. Either add a third placeholder or remove the "Hint 3" reference from the spec text.

```suggestion
<details>
<summary>Hint 2</summary>
{Address a common stumbling point}
</details>

<details>
<summary>Hint 3</summary>
{More direct nudge toward the solution path}
</details>

## Submission
```

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

[greptile-apps]

Context7 MCP dependency not declared in frontmatter

Step 2 instructs the agent to call mcp__plugin_context7_context7__resolve-library-id and mcp__plugin_context7_context7__query-docs, but these MCP tools are not listed in the frontmatter tools: field and depend on the consumer having the Context7 plugin installed. In an IDT consumer without Context7, this step will fail at runtime with no graceful fallback. Consider adding a conditional note ("if Context7 is installed, use …; otherwise fall back to WebFetch") to make the dependency explicit.

Prompt To Fix With AI

This is a comment left during a code review.
Path: agents/research-agent.md
Line: 59-61

Comment:
**Context7 MCP dependency not declared in frontmatter**

Step 2 instructs the agent to call `mcp__plugin_context7_context7__resolve-library-id` and `mcp__plugin_context7_context7__query-docs`, but these MCP tools are not listed in the frontmatter `tools:` field and depend on the consumer having the Context7 plugin installed. In an IDT consumer without Context7, this step will fail at runtime with no graceful fallback. Consider adding a conditional note ("if Context7 is installed, use …; otherwise fall back to WebFetch") to make the dependency explicit.

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