feat(playbooks): add a Claude Haiku 5.5 adaptation chapter

[kyle-sexton]

Closes #4348

Summary

Claude Haiku 5.5 is released with its own prompting guide, so the decision #4348 asks for is: write the chapter. haiku-5-5.md is the playbook's first Haiku chapter. Like its siblings, it is links-only: each entry states this repository's decision in our words, a trigger, and a pointer with an as-of date and a recheck trigger.

Fix

• plugins/playbooks/reference/model-adaptation/haiku-5-5.md: new chapter covering where Haiku runs here, effort, where a run stops, verification, search, thinking, mid-turn messages, JSON/tools/chatbots, migration from Haiku 4.5, system-prompt surfaces, refusals, and caching/cost. It records source byte counts and MD5s.
• plugins/playbooks/skills/fable-5/SKILL.md: meta-rule 3 routes Claude Haiku 5.5 to haiku-5-5.md, and its example of a family with no chapter is now Haiku 4.5. The description lists Haiku 5.5 as current and adds the 'haiku 5.5 adaptation' trigger.
• plugins/playbooks/skills/fable-5/evals/evals.json: two new cases, route-haiku-5-5-session-at-arm-time and trigger-haiku-5-5-adaptation-chapter.
• docs/plugin-philosophy.md: the recheck trigger on the effort-pin rule fired ("a Haiku model appears among the models that support effort"). The rule still says to omit the pin on the haiku rung. Its stated reason changes: the alias resolves to an effort-capable model only on some providers.
• docs/conventions/loop-lane/README.md and CHANGELOG.md (10.2.1, patch): the haiku exclusion stays, but its reason held only where the alias still resolves to the older model. The binding decision is filed as loop-lane: decide whether a capability tier binds haiku now that it resolves to Haiku 5.5 #6856.
• docs/official-docs.md: a row for the Haiku 5.5 prompting guide.
• session-flow observer_analysis_model text (plugin.json, README.md, reference/observer.md): it no longer calls claude-haiku-4-5 "the cheapest active tier", because the pricing page now lists Haiku 5.5 below it. The default is unchanged; whether to move it is session-flow: decide whether the observer's default analysis model moves from claude-haiku-4-5 to claude-haiku-5-5 #6857.
• Changelog fragments: playbooks (minor), session-flow (patch).

Left unchanged on purpose: plugins/multi-agent/reference/defaults.yaml (a separate audit owns it), test fixtures and eval sample answers that use claude-haiku-4-5 as data, and CHANGELOG history.

Verification

Source	Read	Detail
platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-haiku-5-5 (raw .md)	2026-10-10	15,006 B, MD5 f4f01a9a753158e0fd657c7ff4de7a48
platform.claude.com/docs/en/models/haiku-5-5/whats-new-haiku-5-5	2026-10-10	12,830 B, MD5 67e91938ac29b731ef73647ff71b6e1f
platform.claude.com/docs/en/models/haiku-5-5/migration-guide	2026-10-10	19,755 B, MD5 c5f7a69b12cebfac99a7ec02e14f2691
platform.claude.com/docs/en/about-claude/models/overview, choosing-a-model, pricing	2026-10-10	Haiku 5.5 current, Haiku 4.5 legacy
code.claude.com/docs/en/model-config	2026-10-10	116,499 B, MD5 86ee02d08683a5b8696a4b41fb206fef; haiku resolves to Haiku 5.5 on the Anthropic API, Haiku 4.5 on other providers

Checks run locally, all passing:

• markdownlint-cli2 on the 10 changed markdown files: 0 issues.
• scripts/check-changed-skills.sh origin/main: fable-5 PASS (0 errors). Its warnings are about size, and include the description growing from 985 to 1020 characters, still under the 1024 cap.
• scripts/check-skill-description-voice.sh origin/main: PASS.
• check-jsonschema against plugins/skill-quality/reference/evals.schema.json: ok. check-evals-quality.sh: PASS, 0 warnings.
• scripts/check-changelog-fragments.sh --check and --check-required origin/main: pass. scripts/check-changelog-parity.sh --check and --check-bump origin/main: pass.
• scripts/validate-plugins.sh: all manifests valid.
• check-loop-lane-floor-drift.sh, check-conformance-registry.sh and check-docs-naming.sh with --check: pass.

The new evals were not run live (no claude plugin eval run).

Related

• loop-lane: decide whether a capability tier binds haiku now that it resolves to Haiku 5.5 #6856: loop-lane, decide whether a tier binds haiku.
• session-flow: decide whether the observer's default analysis model moves from claude-haiku-4-5 to claude-haiku-5-5 #6857: session-flow, decide whether the observer default moves to claude-haiku-5-5.
• feat(playbooks): apply the Sonnet 5.5 prompting guide across the marketplace #5767: the Sonnet 5.5 chapter this one follows.
• Not done here: widening the audit-instructions and audit-prompting-postures model-scoped rows to haiku-5-5, which the Sonnet 5.5 PR did for Sonnet. playbooks: add a Haiku 5.5 adaptation chapter when Haiku 5.5 ships #4348 does not require it.

🤖 Generated with Claude Code