team-ai-workflows boundary

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

Boundary and evidence

This analysis covers the shipped repository template for distributing team skills and a Markdown knowledge base to Claude Code users. Its intended use is machine-wide access to team skills and retrieval of team knowledge while working in other projects. The target is a memory/knowledge/context-engineering system, not an enclosing agent runtime: its model-dependent portion is the skill workflow served to Claude Code, while shell and Python programs perform link synchronization and lexical retrieval. The README describes that division (SRC-1, README.md):

A template for distributing Claude Code skills and a team knowledge base through a single git repository. --- README.md

The boundary is whole-system for this shipped template. It includes onboarding instructions, installation of user-level skill links and a Git hook, configuration of the knowledge-root environment variable, update-time link maintenance, the retained Markdown corpus, search code, and the prompt that makes the host generate queries and consume retrieved files. The same-repository setup driver and search consumer are included rather than treated as external applications (SRC-1, .claude/skills/team-setup/SKILL.md, .claude/skills/team-setup/scripts/post-merge, .claude/skills/team-setup/scripts/sync_user_skills.sh, skills/docs-grep/SKILL.md, skills/docs-grep/scripts/docs_search.py).

The model-dependent query and answer steps are declared instructions, not behavior implemented by the Python search program (SRC-1, skills/docs-grep/SKILL.md):

Derive 4–8 search keywords from the question. Include synonyms, abbreviations, and both English and native-language variants where relevant. --- skills/docs-grep/SKILL.md

Read the top-ranked files and answer the question. Cite the relative paths of the files you used. If the hits look weak, adjust the keywords and repeat Step 2. --- skills/docs-grep/SKILL.md

The corpus is maintained as repository files rather than through a shipped knowledge-write API. The README instructs adopters to replace samples, and the included writing guidance addresses file authors; the inspected tracked tree contains no separate ingestion service, evaluator or automated knowledge-maintenance runner (SRC-1, README.md, sources/tips/writing-knowledge.md).

Replace the samples in sources/ with your team's knowledge (docs copies, API references, first-hand tips) --- README.md

Coverage of the tracked top level

All tracked text files were inspected. The following table accounts for every top-level tracked file or directory at the frozen commit; no material shipped operational path remains uninspected.

This coverage derives from the frozen tracked tree and SRC-1's paths listed in the table. Source inspection traces /team-setup to its synchronization script and installed hook, and /docs-grep to its search program and subsequent model-side consumption of corpus files. No tests or evaluator are shipped in that tree. No target programs, examples or model calls were executed.

External dependencies and evidence limits

Claude Code supplies skill discovery, model execution, interactive user questions, tool execution and session behavior. Git supplies distribution and hook invocation; the user's shell and filesystem supply links, with Windows-specific utilities used by the synchronization script. Python supplies search execution. These dependencies are identified by shipped instructions and code but their implementations are outside the frozen source. Consequently, this run cannot establish host discovery semantics, actual session restart behavior, operating-system compatibility, successful installation or end-to-end answer quality (SRC-1, README.md, .claude/skills/team-setup/SKILL.md, .claude/skills/team-setup/scripts/sync_user_skills.sh, skills/docs-grep/scripts/docs_search.py). Local user configuration, installed links and actual team deployments are not supplied; their state and operational outcomes cannot be inferred.

The README's external design write-up and the host documentation are deliberately outside the source set. Their arguments, history and host guarantees cannot support this run. The uninspected GIF is a deliberate exclusion, not unavailable evidence. No material source-text access gap was encountered.

The frozen evidence is repository https://github.com/umezy/team-ai-workflows at full commit 9aebd7a6c2803da450904eb5621ae5439508ff94, with applicability cutoff 2026-11-09. The overall tier is code-grounded for shell/Python behavior and doctrine/design for model-facing instructions and declared intent. This is source-level analysis, not observed operation or a causal experiment. Later revisions, the current live repository and external articles are not evidence for this run.

Source register