From ai-writing-plugin
中文优先指导 HARA document type work,同时保留 HARA terminology、evidence、source tier、HITL、final-report、sample/reference 和 candidate-update boundaries。
How this skill is triggered — by the user, by Claude, or both
Slash command
/ai-writing-plugin:haraThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill for `task_type: hara`.
Use this skill for task_type: hara.
默认用中文向用户解释 HARA workflow、材料角色、open confirmations、pending 专业判断和最终 artifact。保留 HARA、hazard、hazardous event、S/E/C、ASIL、safety goal、HITL、NEEDS_USER_CONFIRMATION 等专业术语和机器字段。
如果用户输入或引用证据为英文,可以保留原文片段;说明、提醒和最终回复优先中文。中文表达不能把 HARA candidate wording 写成专业批准。
HARA supports functional-safety hazard analysis report assistance. It helps create a traceable, review-ready HARA package with evidence boundaries, open confirmations, and conservative candidate wording. It must not automatically approve HARA professional judgments.
hara is an official L3 built-in document type backed by built-in DocumentTypeRules, fixtures, regression tests, and this domain guideline.
HARA output may and must preserve HARA terminology: hazard, hazardous event, severity, exposure, controllability, ASIL, S/E/C, risk level, and safety goal.
Typical HARA inputs include:
The HARA default sections are:
HARA critical claims include:
These claims require source evidence or HITL confirmation. Without evidence or a recorded human decision, keep NEEDS_USER_CONFIRMATION.
HARA professional judgment cannot be finalized by AI alone. The following require qualified human review and recorded HITL confirmation:
User confirmation must be recorded in trace/hitl_decisions.jsonl or an equivalent trace artifact.
Without sufficient T0/T1 support and explicit HITL, do not write final claims such as:
These are warning examples, not recommended output.
source materials can support project facts when parsed and relevant.sample is not fact source. sample must not be used as fact source. sample HARA reports may guide format, style, table shape, and wording style only. reference must not prove project-specific facts. reference cannot prove project facts. T3/T4/T5 cannot support critical claim by themselves.
Review focus includes:
Verification focus includes:
NEEDS_USER_CONFIRMATION preservationThe HARA final_report.md is a plugin-generated final package artifact for human review. final report is not approval, not formal compliance approval, and not qualified safety sign-off.
Valid conservative statuses include finalized_with_open_items, ready_for_human_review, and blocked_pending_confirmation. Unresolved HARA items must remain open.
.venv/bin/python -m ai_writing_plugin write-run --task examples/hara_demo_fixture/task.yaml
NEEDS_USER_CONFIRMATION without real HITL.candidate_profile_update.yaml and candidate_skill_patch.md must not overwrite a stable skill.npx claudepluginhub exquisite0828/ai_writer_plugin --plugin ai-writing-pluginGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.