Rosetta Add Framework Docs

Finalise + commit + PR

Last phase of rosetta-add-framework. Updates docs, commits, raises a PR.

Inputs

• FRAMEWORK — slug (e.g. google-adk)
• FRAMEWORK_DIR — directory name (e.g. google-adk-py)
• LANGUAGE — py or ts
• FRAMEWORK_DISPLAY — human-readable name (e.g. "Google ADK")
• FRAMEWORK_HOMEPAGE — URL for the README link (e.g. https://google.github.io/adk-docs/)
• MODEL_OVERRIDE (optional) — note if the framework needs a non-default Claude model
• SESSION_TAGGING — auto | via-using-session — for the PR description

Steps

1. README — supported-frameworks table

Edit /Users/jimbobbennett/github/project-rosetta-stone/README.md. Find the Supported frameworks table and insert a new row alphabetically.

| [<FRAMEWORK_DISPLAY>](<FRAMEWORK_HOMEPAGE>) | ✅ | — |

(Use ✅/— based on LANGUAGE. Python frameworks: ✅ in Python column, — in TypeScript. Vice versa for TS.)

2. README — directory tree

Find the three tier blocks under "## What's in the box". Insert the new framework alphabetically in each:

│   ├── <FRAMEWORK_DIR>/         <FRAMEWORK_DISPLAY> (Python + Next.js)

(Indent and spacing must match the surrounding lines — copy an existing line and edit.)

3. README — Frameworks comparison table

Find the Frameworks comparison table. Insert a new row alphabetically:

| **<FRAMEWORK_DISPLAY>** | <agent library> | <LLM client> | <streaming API> | Python FastAPI backend + Next.js frontend |

Source the four cell values from the research phase notes. Keep concise — one phrase per cell, code-fenced for class/function names.

4. README — "what differs between tiers"

Find "## What changes between tiers?" and insert a new section alphabetically:

For **<FRAMEWORK_DISPLAY>**, only these files differ:

- `backend/tracing.py` — tracing initialization (new file, imported before `<framework>`)<inline note about any quirk: register vs manual TracerProvider, instrument_all, etc.>
- `backend/main.py` — imports `backend.tracing` before other backend modules
- `backend/requirements.txt` — observability packages (`arize-phoenix-otel` or `arize-otel` + `openinference-instrumentation-<framework>`)
- `env.example` — observability environment variables

5. Orchestrator TODO snapshot

Edit /Users/jimbobbennett/github/project-rosetta-stone/.claude/skills/rosetta-add-framework/SKILL.md. Find the framework's [ ] line in the TODO section and change it to [x]. Update the "last refreshed" date at the top of the TODO section.

6. Save framework gotchas to memory

For each non-obvious thing discovered (model override, manual tracer-provider, instrument_all, session caveat, eval extractor mismatch), save a memory entry:

• type: project
• name: framework_<framework>_<aspect> (e.g. framework_crewai_strict_tools, framework_pydantic_ai_instrument_all)
• body: lead with the fact, then Why: and How to apply: lines per the auto-memory <body_structure> rules

These accumulate over time as more frameworks are added — making each new addition faster.

7. Commit per tier + push + PR

cd /Users/jimbobbennett/github/project-rosetta-stone
BRANCH="feat/add-<FRAMEWORK>"
git checkout -b "$BRANCH"

# Stage and commit each tier separately so reviewers see the progression
git add no-observability/<FRAMEWORK_DIR>/
git commit -m "$(cat <<EOF
Add <FRAMEWORK_DISPLAY> — no-observability tier

<one paragraph: framework API choice, streaming approach, any model override>

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EOF
)"

git add phoenix/<FRAMEWORK_DIR>/
git commit -m "$(cat <<EOF
Add <FRAMEWORK_DISPLAY> — Phoenix tier

<one paragraph: tracing pattern, any session.id wrap>

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EOF
)"

git add ax/<FRAMEWORK_DIR>/
git commit -m "$(cat <<EOF
Add <FRAMEWORK_DISPLAY> — AX tier

<one paragraph: tracing pattern>

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EOF
)"

git add README.md .claude/skills/rosetta-add-framework/SKILL.md
# Stage Playwright setup if this was the first framework to add it
[ -f playwright/package.json ] && git status --short playwright/ | grep -q '?? playwright/' && git add playwright/

git commit -m "$(cat <<EOF
Document <FRAMEWORK_DISPLAY> in README + mark off in TODO

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EOF
)"

git push -u origin "$BRANCH"

8. Open the PR (initial body, screenshots appended in step 9)

gh pr create --title "Add <FRAMEWORK_DISPLAY>" --body "$(cat <<EOF
## Summary

- New tiers: \`no-observability/<FRAMEWORK_DIR>/\`, \`phoenix/<FRAMEWORK_DIR>/\`, \`ax/<FRAMEWORK_DIR>/\` running the identical Wonder Toys agent
- README updated: supported-frameworks table, directory tree, frameworks comparison, what-differs section
- TODO marked off in the rosetta-add-framework orchestrator skill
- <any framework-specific notes: model override, manual tracer-provider, session.id wrap, etc.>

## Verification

- [x] no-observability: lint + smoke chat + multi-turn history confirmed
- [x] phoenix: trace lands in \`wonder-toys-<framework>\` project; 25 synthetic requests + evals harness green
- [x] ax: trace lands in configured AX project; 25 synthetic requests green
- [x] session.id: <auto-tagged | tagged via \`using_session(user_id)\` wrap in agent.py>
- [x] Playwright public-flow smoke green
- [x] Screenshots attached (appended below by rosetta-pr-screenshots)

🤖 Generated with [Claude Code](https://claude.com/claude-code)
EOF
)"

9. Capture and attach screenshots

After the PR opens, invoke the rosetta-pr-screenshots skill to:

• Run a canned 3-turn conversation through the AX and Phoenix backends
• Screenshot the session view + each trace in both AX and Phoenix
• Screenshot the Wonder Toys app UI via Playwright (public pages)
• Upload all images as a GitHub release asset (screenshots-pr-<N>)
• Append a ## Screenshots section to the PR body referencing the asset URLs

Inputs to pass:

• FRAMEWORK — same slug used here
• FRAMEWORK_DIR — same dir name
• PR_NUMBER — from the gh pr create output
• AX_PORT / PHOENIX_PORT / NEXT_PORT — match the ports the orchestrator assigned for this run
• BRANCH — the feature branch we just pushed

The skill is idempotent — safe to re-run if screenshots need refreshing.

Output

PR opened: <URL>
  branch: feat/add-<FRAMEWORK>
  commits: 4 (no-obs, phoenix, ax, docs)
  README updated: supported-frameworks, tree, frameworks-table, what-differs
  TODO marked off in orchestrator skill
  Memory saved: <count> framework gotchas
  Screenshots attached: <count> AX + <count> Phoenix + <count> UI (via release <tag>)