Epic: Agentic Bridge — Cross-Stitch ADP Catalog × Skills × Articles

[julianken]

Overview

The Agentic Bridge cross-stitches three on-site silos at detached-node.dev — the 23-pattern ADP catalog, the 8 active skills, and the 3 published process articles — so each pattern resolves into a single readable triple for next-gen agentic devs:

1. Abstract pattern (the ADP satellite content)
2. Realization in Claude Code primitives + structural scaffolding (with full GitHub URLs to PR/issue/doc evidence)
3. Reader's move — one concrete action with measurable signal

The teaching layer ships as a 3rd collapsible disclosure on each ADP satellite (after Overview + Background), plus 2 new Layer-5 satellites, 1 new bridge essay, 4 cross-link sub-blocks on existing satellites, and a Wave-0 cleanup that closes paper-vs-practice gaps.

Audience: someone who already uses Claude Code and modifies their setup (writes skills, edits subagents, configures hooks). The bridge teaches how to apply the patterns; novelty is not a goal of the bridge content.

Wave structure

Wave	Hours	PRs	Goal
W0 — Cleanup	2	2	Restore .claude/hooks/, remove worktree-agent-* dead code, fix nonexistent script refs
W1 — Foundation + Spine	~24	5	Extend Pattern type + render component + Vitest validator + 4 host-satellite Tier A/B authorings
W2 — Continuation + Satellites + Essay	~47	9	5 Tier A/B continuations, 2 promote-to-ADP satellites, 4th bridge essay, 4 leave-in-bridge Tier C cross-links
W3 — Umbrella + Honest-absence + Long tail	~13	4	12-factor Tier U umbrella, Reflexion+Streaming honest-absence, 9 skill-only Tier C pointers, advisory realizes: frontmatter pass

MVP = Wave 1 (~24h, 5 PRs). Full publication = waves 0-3 (~86h total).
Calendar estimate: 4-7 weeks for waves 0-2 / 8-10 weeks for full publication.

Locked decisions

• Linear has been completely scrubbed from all 5 active skills. GitHub Issues is the source of truth.
• Memory Management Monday test uses wc -w × 1.8 proxy or tiktoken (cl100k_base) canonical. The earlier × 1.33 multiplier was empirically off by 27%.
• Q1 (promote-to-ADP timing) = Wave-2 reading: satellites land alongside the 4th essay in Wave 2.
• Anthropic April 2026 multi-agent code review is structurally identical to identity-separated review and is absorbed in the new /identity-separated-review satellite framing as the managed-feature equivalent path.
• scripts/verify_bridge_urls.sh is a Wave 3 deliverable — every full GitHub URL anchor in bridge content must resolve.
• Wave 1 omits [draft] cross-links entirely — anchors light up only when the target lands. Mitigates Wave-2 stall risk.
• 4th bridge essay slug = phase-boundary-loop.
• identity-separated review companion article title = "Cross-Identity Code Review".
• analysis-funnel mirror scope = references/ + scripts/ from ~/.claude/skills/analysis-funnel/.

Acceptance criteria (Epic-level)

• All 24 child issues carry bot:approved label
• docs/agentic-bridge/reframe/analysis-v1.md and action-plan-v1.json haven't drifted at start-of-execution time
• Wave 0 PRs merged (W0.1 + W0.2) before Wave 1 PRs open
• Wave 1 ships without [draft] cross-link anchors
• All linked GitHub URLs in bridge content resolve via scripts/verify_bridge_urls.sh

Linked issues

Wave 0 — Cleanup (~2h)

• [W0.1] Restore PreToolUse hooks + remove worktree-agent dead code · [W0.1] Restore PreToolUse hooks + remove worktree-agent dead code #305
• [W0.2] Fix nonexistent-script refs + restore analysis-funnel mirror dirs · [W0.2] Fix nonexistent-script refs + restore analysis-funnel repo-mirror dirs #306

Wave 1 — Foundation + Spine (~24h)

• [W1.0] Pattern type extension + RealizingDisclosure component + Vitest validator · [W1.0] Add realizingInClaudeCode to Pattern type + RealizingDisclosure component + Vitest validator #307
• [W1.1] Orchestrator-Workers Realizing-in-CC (Tier A) · [W1.1] Orchestrator-Workers Realizing-in-CC (Tier A) #308
• [W1.2] Context Engineering Realizing-in-CC (Tier A) + minimal-CLAUDE.md inline Tier C · [W1.2] Context Engineering Realizing-in-CC (Tier A) + minimal-CLAUDE.md inline Tier C #309
• [W1.3] Evaluation (LLM-as-Judge) Realizing-in-CC (Tier A) · [W1.3] Evaluation (LLM-as-Judge) Realizing-in-CC (Tier A) #310
• [W1.4] Memory Management Realizing-in-CC (Tier B) · [W1.4] Memory Management Realizing-in-CC (Tier B) #311

Wave 2 — Continuation + New satellites + 4th essay (~47h)

• [W2.1] Parallelization Realizing-in-CC (Tier A) · [W2.1] Parallelization Realizing-in-CC (Tier A) #312
• [W2.2] Checkpointing Realizing-in-CC (Tier A) · [W2.2] Checkpointing Realizing-in-CC (Tier A) #313
• [W2.3] Code Agent Realizing-in-CC (Tier B) + worktree-per-issue inline Tier C · [W2.3] Code Agent Realizing-in-CC (Tier B) + worktree-per-issue inline Tier C #315
• [W2.4] Guardrails Realizing-in-CC (Tier A) with PR chore(skills): refresh repo .claude/ + strip Vercel MCP allowlist #287 anti-example · [W2.4] Guardrails Realizing-in-CC (Tier A) with PR #287 anti-example footnote #316
• [W2.5] MCP Realizing-in-CC (Tier B) · [W2.5] MCP Realizing-in-CC (Tier B) #317
• [W2.6] identity-separated-review (NEW Layer-5 satellite) + "Cross-Identity Code Review" article · [W2.6] identity-separated-review (NEW Layer-5 satellite) + scoper rank-1 article #318
• [W2.7] funnel-method (NEW Layer-5 satellite) + 4th bridge essay (slug phase-boundary-loop) · [W2.7] funnel-method (NEW Layer-5 satellite) + 4th bridge essay (teaching voice) #319
• [W2.8] tool-use-react Tier C cross-link (worktree hook half) · [W2.8] tool-use-react Tier C cross-link (worktree hook half) #320
• [W2.9] Four leave-in-bridge Tier C cross-link sections (Mergify queue + 5-section PR body) · [W2.9] Four leave-in-bridge Tier C cross-link sections (3 hosts) — Mergify queue + 5-section PR body #321

Wave 3 — Umbrella + Honest-absence + Long tail (~13h)

• [W3.1] 12-factor-agent umbrella Tier U + token-economics canonical home · [W3.1] 12-factor-agent umbrella Tier U + token-economics canonical home #322
• [W3.2] Reflexion + Streaming honest-absence Tier D pointers · [W3.2] Reflexion + Streaming honest-absence Tier D pointers #324
• [W3.3] Nine skill-only patterns Tier C pointers · [W3.3] Nine skill-only patterns Tier C pointers #325
• [W3.4] realizes: frontmatter advisory pass on 5 active SKILL.md · [W3.4] realizes: frontmatter advisory pass on 5 active SKILL.md #326

Tooling (Wave 3)

• [TOOL-1] scripts/verify_bridge_urls.sh · [TOOL-1] Tooling: scripts/verify_bridge_urls.sh #327
• [TOOL-2] scripts/check_meta_rule_consistency.sh · [TOOL-2] Tooling: scripts/check_meta_rule_consistency.sh #328
• [TOOL-3] SKILL.md frontmatter parser unknown-key tolerance test · [TOOL-3] Tooling: SKILL.md frontmatter parser unknown-key tolerance test (one-shot) #329

Recurring post-publication ops (umbrella)

• [POST-OPS] Recurring post-publication followups · [POST-OPS] Recurring post-publication followups (umbrella) #330

References

• Analysis: docs/agentic-bridge/reframe/analysis-v1.md
• Action plan: docs/agentic-bridge/reframe/action-plan-v1.json
• Original funnel report: docs/agentic-bridge/funnel/phase-4/analysis-report.md
• Citation discipline applies throughout: configurations are described as "what worked here," not authorship claims.