feat: Social media campaign — harness design launch (4-week arc)

[ryaneggz]

Social Media Campaign — Harness Design Launch

A 4-week marketing campaign tied to Anthropic's "Harness design for long-running application development" publication. Positions Ruska AI / Orchestra as the open-source platform shipping this research as usable infrastructure.

Campaign Structure

Week	Theme	Platforms
1	Signal Post — React to Anthropic's article, position Orchestra	LinkedIn + X/Twitter
2	Build-in-Public — Show evaluator agent implementation learnings	LinkedIn + X/Twitter
3	Demo Post — Showcase full harness run with results	LinkedIn + X/Twitter
4	Thought Leadership — Lessons and vision for AI engineering	LinkedIn + X/Twitter

Key Dependencies

• Week 2 requires evaluator agent work in progress (issue/PR to reference)
• Week 3 requires a working demo of planner + generator + evaluator loop
• Week 3 requires cost tracking to report run costs

Deliverables

• Week 1: LinkedIn post + X/Twitter 6-tweet thread (within 48hrs of article)
• Week 2: LinkedIn post + X/Twitter 4-tweet thread (build-in-public)
• Week 3: Demo video + LinkedIn post + X/Twitter post
• Week 4: Thought leadership LinkedIn post + X/Twitter post
• UTM tracking links configured
• Engagement in first 2 hours of each post

Spec & Plan

• Full spec: .claude/specs/social-media-campaign.md
• Execution plan: .claude/plans/social-media-campaign.md

[ryaneggz]

Feature Dependency Map for Campaign Timing

Analysis of which features must ship (or be in-progress) before each week's content can be credibly created and posted.

---

Week 1: The Signal Post (March 24-26, 2026)

Hard dependencies: NONE

This is a pure reaction post to Anthropic's article. The content references what Orchestra will ship, not what is already working. It links to existing GitHub issues as proof of commitment.

Soft dependencies (for credibility):

• Open issues with specs for evaluator (feat: Evaluator Agent — GAN-inspired generator/evaluator feedback loop #899), planner (feat: Planner Agent — auto-expand short prompts into full product specs #898), and harness templates (feat: Harness Templates — reusable planner→generator→evaluator architectures #911) — these already exist and are linkable today
• The repo must be public and presentable (README, existing feature set)

Verdict: Can ship immediately upon article publication. No feature work blocks this.

---

Week 2: The Build-in-Public Post (~March 31 - April 2, 2026)

Hard dependencies:

• Evaluator Agent (feat: Evaluator Agent — GAN-inspired generator/evaluator feedback loop #899) — must be actively in-progress. The post shares specific implementation learnings (schema design, skepticism prompting, few-shot calibration). At minimum, the EvaluationCriterion / EvaluatorConfig / EvaluationResult schemas and the evaluator system prompt need to exist in a PR branch. The post explicitly references the schema (EvaluationCriterion: name, description, weight, threshold) and implementation decisions.

Not required:

• Evaluator does not need to be merged/shipped — an open PR with real code is sufficient (and ideal for build-in-public framing)
• Planner agent, harness templates, and cost tracking are not referenced in Week 2 content

Verdict: Requires ~1 week of evaluator agent development started by Week 1.

---

Week 3: The Demo Post (~April 7-9, 2026)

Hard dependencies (ALL must be working end-to-end for a real demo):

1. Planner Agent (feat: Planner Agent — auto-expand short prompts into full product specs #898) — must produce a structured spec from a short prompt (the demo shows "14 features across 8 sprints" expanded from one sentence)
2. Evaluator Agent (feat: Evaluator Agent — GAN-inspired generator/evaluator feedback loop #899) — must run the generate-evaluate-critique-retry loop (the demo shows "3 evaluation rounds" catching "3 critical bugs")
3. Cost Tracking (feat: Per-run cost tracking, phase-level metering, and budget caps #905) — must report per-run cost (the demo claims "Total cost: ~$85")
4. Harness Templates (feat: Harness Templates — reusable planner→generator→evaluator architectures #911) — referenced in the plan as a Week 3 dependency; the "Full Harness" template bundles planner + evaluator together

Supporting dependencies:

• Context Reset (feat: Context reset with structured handoff artifacts for long-running sessions #906) — the harness templates spec references context_strategy field; the "Full Harness" template uses context_strategy: "auto". Needed if the demo runs long enough to trigger resets.
• Playwright MCP integration — the evaluator spec mentions Playwright-based UI testing. The demo post describes the evaluator catching visual bugs ("Mobile viewport was completely broken below 768px"), implying Playwright interaction.

Risk mitigation note from the plan: "Use staging/development environment for the demo. The demo does not require production-shipped features, only a working end-to-end run." This means features can be in late-stage PR state, not necessarily merged to development.

Verdict: This is the critical-path week. Four features (#898, #899, #905, #911) must be functional (even if unmerged). Requires ~2 weeks of focused development from campaign start.

---

Week 4: Thought Leadership (~April 14-16, 2026)

Hard dependencies:

• At least one feature shipped to production — the plan explicitly states this. The post says "After shipping evaluator agents, planner agents, and harness templates in Orchestra..." Credibility requires at minimum evaluator or planner merged to development.
• Weeks 1-3 posts published — the thought leadership synthesizes the campaign arc

Soft dependencies:

• Engagement data from Weeks 1-3 (to refine messaging)
• Ideally multiple features shipped, but the post can be written around whatever has landed

Verdict: Requires at least evaluator (#899) or planner (#898) merged to development by ~April 14.

---

Summary: Critical Path

Week 1 (Mar 24-26)  No blockers — ship immediately
     |
     v
Week 2 (Mar 31)     Evaluator (#899) in-progress (PR with schemas + prompt)
     |
     v
Week 3 (Apr 7-9)    Planner (#898) + Evaluator (#899) + Cost Tracking (#905)
                     + Harness Templates (#911) — all functional in staging
     |
     v
Week 4 (Apr 14-16)  At least 1 feature merged to development

Features That Could Start Early (Parallel Work)

Feature	Can start before campaign?	Notes
Evaluator Agent (#899)	Yes — start NOW	Needed by Week 2; longest lead time
Planner Agent (#898)	Yes — start NOW	Independent of evaluator; needed by Week 3
Cost Tracking (#905)	Yes — start NOW	Independent; needed for Week 3 demo metrics
Harness Templates (#911)	Start after #898 + #899 have schemas	Depends on PlannerConfig + EvaluatorConfig schemas
Context Reset (#906)	Can defer	Only needed if Week 3 demo is a long-running session
Observability (#904)	Can defer	v0.10.0 milestone; not referenced in any campaign content
Sprint Contracts (#910)	Can defer	Not referenced in campaign content

Recommendation

Start #899 (Evaluator), #898 (Planner), and #905 (Cost Tracking) in parallel immediately. These three are independent of each other and all feed into the Week 3 demo. Harness Templates (#911) can begin once the evaluator and planner schemas stabilize (~1 week in). Week 1 content can be posted today with zero feature dependencies.

[ryaneggz]

Dependency Graph Validation

Position in Dependency Graph

Social Media Campaign (#902) — CONSUMER (Layer 2, non-blocking)
  └─ depends on: Evaluator (#899), Planner (#898), Cost Tracking (#905)
  └─ depended upon by: (nothing)

Declared Dependencies

• Week 2 requires evaluator agent work in progress
• Week 3 requires a working demo of planner + generator + evaluator loop
• Week 3 requires cost tracking to report run costs

Validated Dependencies

Correct, but the timeline implications are significant. The campaign is a 4-week arc starting "within 48 hours" of the Anthropic article (March 24, 2026 — today's date per context). This means:

Week	Date	Dependency	Status
Week 1	Mar 24-26	None (reaction post)	Can ship now
Week 2	~Mar 31	Evaluator (#899) in progress	Evaluator must be started
Week 3	~Apr 7	Planner + Evaluator + Cost Tracking all working	All 3 must be shipped
Week 4	~Apr 14	None (thought leadership)	No deps

Implicit Dependencies Found

1. Demo infrastructure — Week 3 requires "a working demo" and a "demo video." This implies a stable deployment environment where the planner-generator-evaluator pipeline runs end-to-end. There is no spec covering demo/staging environment setup.
2. Cost reporting UI — Week 3 references reporting "run costs" (~$85). This requires the Cost Tracking (feat: Per-run cost tracking, phase-level metering, and budget caps #905) frontend cost badge to be visible, not just the backend metrics capture.
3. UTM tracking — The spec mentions "UTM tracking links configured" as a deliverable but no spec or infrastructure covers UTM configuration.

Interface Contracts Needed

None — this is a content/marketing spec, not a code spec. However, it should define:

• What constitutes a "working demo" (specific user flow, which template, what prompt)
• Whether the demo is against production, staging, or local

Risk Assessment

• Timeline risk is high. Week 3 (~April 7) requires THREE specs to be fully shipped (Planner, Evaluator, Cost Tracking). Given that none have started and each has ~5 implementation phases, this is extremely aggressive.
• Recommendation: Decouple Week 3 from Cost Tracking. Report costs manually from API/logs instead of requiring the full frontend to be ready. Alternatively, push Week 3 to Week 4-5.

Critical Path Assessment

Not on critical path for code. This is a marketing deliverable that consumes features from other specs. However, it provides the strongest external pressure to ship Planner + Evaluator quickly. Week 1 content can ship immediately with no code dependencies.

[ryaneggz]

Phased Implementation Roadmap

Based on analysis from 20 specialized subagents studying the codebase, here is the implementation plan across all harness design features.

Phase 0: Shared Foundation (Prerequisites) — #914

Land on development before feature branches start.

• Shared AgentPhase enum (chore: Phase 0 — shared foundation prerequisites for harness design features #914)
• Frontend "custom" SSE event plumbing (chore: Phase 0 — shared foundation prerequisites for harness design features #914)
• Frontend HITL interrupt UI (chore: Phase 0 — shared foundation prerequisites for harness design features #914)
• Model context window mapping (chore: Phase 0 — shared foundation prerequisites for harness design features #914)

Phase 1: Core Infrastructure (parallel, no dependencies)

• Cost Tracking (feat: Per-run cost tracking, phase-level metering, and budget caps #905) — extend existing cache_metrics_middleware, add pricing table, budget caps
• Planner Agent (feat: Planner Agent — auto-expand short prompts into full product specs #898) — pre-execution phase in LLMController, /PLAN.md persistence, HITL approval

Phase 2: Quality & Resilience (parallel, Phase 1 not blocking)

• Evaluator Agent (feat: Evaluator Agent — GAN-inspired generator/evaluator feedback loop #899) — post-generation controller loop, separate context, grading criteria
• Context Reset Handoff (feat: Context reset with structured handoff artifacts for long-running sessions #906) — extend CompactingMiddleware, HandoffArtifact, session tracking

Phase 3: Observability (after Phase 1, shares infrastructure with Cost Tracking)

• Agent Observability (feat: Agent phase observability — structured traces with frontend timeline #904) — structured spans, agent_spans table (shared with feat: Per-run cost tracking, phase-level metering, and budget caps #905), frontend timeline, optional OTel export

Phase 4: Composition (after Phases 1+2)

• Sprint Contracts (feat: Sprint contracts — agents negotiate scope and pass/fail criteria before execution #910) — requires Evaluator (feat: Evaluator Agent — GAN-inspired generator/evaluator feedback loop #899), benefits from Planner (feat: Planner Agent — auto-expand short prompts into full product specs #898)
• Harness Templates (feat: Harness Templates — reusable planner→generator→evaluator architectures #911) — requires Planner (feat: Planner Agent — auto-expand short prompts into full product specs #898) + Evaluator (feat: Evaluator Agent — GAN-inspired generator/evaluator feedback loop #899), bundles all configs

Phase 5: Launch

• Social Media Campaign (feat: Social media campaign — harness design launch (4-week arc) #902) — Week 1 can ship immediately; Weeks 2-4 follow feature delivery

---

This issue is in Phase 5: Launch.
Week 1 can start immediately. Weeks 2-4 require Evaluator (#899) and Planner (#898) to be in-progress or shipped.