Interview Doc Agent boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

The target is the shipped Interview Doc Agent skill, its persistent experience-library structure, templates and usage instructions. Its intended use is to turn personal experience materials into reusable project dossiers, then generate resumes and interview scripts, optionally tailored to a job description. It qualifies as a memory/knowledge/context-engineering system because model-directed instructions maintain retained dossiers and indexes and later select those dossiers for document generation (SRC-1, SKILL.md, wiki/index.md). It is not an enclosing runtime: the supplied skill delegates execution and file access to a host AI.

由读到本 skill 的 AI 直接执行（Obsidian Claudian / 飞书智能伙伴 / Claude Code 均可），直接读写文件。无需 API key。 --- SKILL.md

The functional boundary includes initialization, user-supplied materials and job descriptions, dossier organization, index maintenance, selective rereading, template-guided generation, and output placement. The maintenance-to-consumption loop is shipped as instructions rather than executable application code (SRC-1, SKILL.md).

用户新增/修改 materials/ 原始素材后：读新文件 → 提炼整合进 经历库/ 对应档案 → 更新 00-个人档案.md 索引和 wiki/index.md 检索表。 --- SKILL.md

• 生成时先读 wiki/index.md 这张轻量索引表定位，再只读相关经历库档案 + 模板，不必全读（省 token） --- SKILL.md

The boundary kind is whole-system for this shipped prompt-and-document artifact, not for its external hosts. All tracked files were inspected, including the empty output placeholder. No shipped executable driver, evaluator, test suite or provider integration appears in the tracked tree. Therefore the evidence tier is doc-grounded: instructions and examples establish the intended loop, not successful execution, reliable data preservation, privacy guarantees or measured token savings (SRC-1, README.md, SKILL.md, docs/01-理念-karpathy-wiki.md, docs/04-使用指南.md). The experience example explicitly identifies its data as fictional, so it is not an observed execution artifact (SRC-1, 经历库/01-示例经历.md).

这是一篇示例经历档案，展示经历库该长什么样。数据为虚构演示。 --- 经历库/01-示例经历.md

External dependencies are a model-capable host with conversation and read/write access, its authentication and permissions, and the local or cloud storage it exposes. Obsidian/Claudian/Claude Code installation and Feishu adaptation instructions are included; the implementations of those products, lark-cli, model providers, and linked Karpathy and resume-template sources are excluded. This prevents conclusions about host instruction loading, tool enforcement, actual provider data handling, cloud persistence or synchronization, and the correctness of external design claims (SRC-1, README.md, docs/02-安装-obsidian.md, docs/03-接入-飞书.md). Feishu writeback is explicitly permission-dependent and may require manual copying; it is not treated as a shipped automatic integration.

飞书智能伙伴能否「写回」知识库取决于版本权限。若不能自动写回，它会把整理好的内容输出在对话里，你手动粘进知识库页面即可。 --- docs/03-接入-飞书.md

The frozen source is the acquired Git checkout at 6ebfe9aba989d7949dadea0fc4af6fb64892c6aa, verified as its HEAD. The applicability cutoff is 2026-10-14. Links to moving download URLs in source instructions do not expand this evidence boundary. There are no unavailable repository files or material paths left uninspected. Actual user dossiers, generation outputs, execution traces and comparative experiments are not supplied; their absence prevents operational and causal conclusions, not inspection of the shipped artifact.

Tracked top-level coverage

Every row below is included and inspected. Directory rows enumerate the material tracked paths. Each source reference uses SRC-1 at the frozen revision.

Source register