Remake the guide for the v0.18.2 era: Part 27 "Power Secrets", Part 28 "Recipe Book", six new infographics + July 2026 research pass

[OnlyTerp]

Summary

Full modernization from the 2026-07-09→17 X/social research and the 2026-07-16 web research briefs — content, structure, and artwork.

Two new flagship parts:

• Part 27 — Power Secrets: a field manual of 25 verified non-obvious mechanics distilled from the official Wingtips series (Update LightRAG setup: Kimi 2.5 entity extraction & Cerebras+Qwen3 options #1–Fix docs for current Hermes gateway, sandbox, and MCP flows #22) plus the highest-signal community findings — context/cache mechanics (memory snapshot rule, compression internals, compaction's structured brief), cost mechanics (gateway token tax, credential pools vs fallbacks, model-switch cache resets, MoA turn economics), profiles/identity architecture, and operational traps (Kanban toolset/workspace, state.db, linger, export --redact) — ending in a printable one-page cheat sheet.
• Part 28 — The Recipe Book: twelve end-to-end production builds observed in the wild (morning finance loop, staged Gmail, SMB approval office, overnight Kanban, GPU FIFO, content swarm, carousel factory, job-hunt pipeline, accountability coach, health correlator, Blender room, secretary office), each mapped to real primitives with its approval posture and known trap.

Six new house-style SVG infographics (matching the existing dark-purple design system): Part 27 hero, the 5×5 cheat-sheet poster, the seven-layer security stack, the seven-rung agent ladder, the gateway token tax, and the Recipe Book map — plus a refreshed release timeline.

Version currency: badges/intro/What's New now target v0.18.2 (v2026.7.7.2) — v0.18.1 patch-rollup note, WhatsApp/Baileys fix with Docker-tag guidance (Part 15), "main → v0.19.0 = experimental" framing, Hermes Cloud labelled preview (Part 24).

Major section additions across 18 existing parts: MCP CVE table + 7-point hardening checklist (17); seven-layer security model, action ontology, external spend kernels (19); gateway token tax, benchmark-the-stack, Langfuse v4/OTEL (20); agent ladder + profiles-as-rooms (8); Kimi K3 / Sol-Terra-Luna landscape, Anthropic-subscription limitation, credential pools (9); compression/compaction internals + rule-file context tax (6); memory snapshot rule + approval gates + memory-stack table (7); LightRAG 2026 scaling playbook (3); skill-description hygiene (5); Kanban traps + overnight/GPU-FIFO shapes (23); /steer///queue///busy semantics (14); coding-agent lane updates (18); sandbox decision guide (21); computer use + current local models (25); MoA measured economics (26); disk/prune/linger ops (11/16); ECOSYSTEM community radar.

Community/X claims are labelled as community signals throughout; all new config snippets stay within the real v0.18 schema (no invented keys). Translations (README-zh/ja) intentionally left at their 2026-07-03 sync (noted in CHANGELOG).

Type

• Docs / content update
• New skill (skills/)
• New config template (templates/config/)
• Benchmark addition
• Ecosystem entry
• Infra template (compose / caddy / systemd / script)
• Fix / typo / link

Checklist

• Cross-links are relative (./partN-foo.md) and resolve — check_anchors.py passes locally
• No secrets in any example — ${VAR} placeholders only
• Dates / prices / PR numbers are current (or marked with the date)
• For skills: N/A (no skill changes; validate_skills.py still passes)
• For templates: N/A
• CHANGELOG.md updated if user-facing

Screenshots / diffs (optional)

New artwork (all hand-built SVG in the guide's existing design language):

Link to Devin session: https://app.devin.ai/sessions/8d8545a09b0a4a24823533ffc96dc573
Requested by: @OnlyTerp

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

Devin Review found 1 potential issue.

View 2 additional findings in Devin Review.

[devin-ai-integration]

🟡 Guide gives two conflicting values for when context compression starts

The guide states the compression trigger is "around 50% of the context window" (part27-power-secrets.md:30) while the same guide elsewhere states the default trigger is ~80%, so a reader following one section will configure or expect the wrong compression behavior.

Impact: Readers get contradictory guidance about when the agent starts summarizing, leading to confusion and mis-tuned compression settings.

Where the numbers disagree

Secret #2 asserts as fact that automatic compression "kicks in around 50% of the context window" (part27-power-secrets.md:30), and the new Part 6 section says "community reports put the practical trigger around the half-full mark on default settings" (part6-context-compression.md:78). But the unchanged Part 6 text directly above still says "Default: when context reaches ~80% of the model's window" with compression.threshold 0.8 (part6-context-compression.md:61, part6-context-compression.md:68-69), and Part 11 references "the 0.8 default". The new content and the surrounding text contradict each other on the single most important compression number.

---

Was this helpful? React with 👍 or 👎 to provide feedback.

[devin-ai-integration]

Good catch — reconciled. Both Part 27 Secret #2 and Part 6 now state the configured default (compression.threshold: 0.8, ~80%) as the source of truth, with the ~50% figure explicitly framed as a community observation explained by fixed overhead (tool definitions + rule files) counting against the window.