DevLoop — business-analysis workflow for this repo

When asked to define a feature, gather requirements, or create stories/tickets, follow this gated chain. Each phase gates the next; do not skip ahead. Each phase has an Agent Skill (installed under .agents/skills/) that carries the full method — adopt it (Codex auto-selects by description, or invoke $<skill> / pick it from /skills). This file only sequences them.

0. context-librarian skill — Act as a Context Librarian. Ask, one category at a time, where company knowledge lives (local files/folders, wikis & URLs, SharePoint, paste-in text). Compile sources into an interlinked LLM-Wiki under knowledge/ (concept articles with [[wikilinks]] + source-traceable frontmatter + an index.md routing layer). Run FIRST when source material exists. 0b. business-analyst skill (optional — waterfall/regulated/SI teams) — produce a Business Requirements Document (brd.md): business objectives + metrics (OBJ-n), scope, stakeholders, current/future state, and numbered business requirements (BR-n) in business language (no "system shall"), with sign-off. Agile teams skip this and go straight to requirements. When a BRD exists, later phases trace BR → FR → US.
1. requirements-analyst skill — Read knowledge/wiki/index.md if present (and brd.md if it exists — tag each FR/NFR with the BR-n it serves), open relevant concept articles, and only ask about gaps. Interview Socratically, ONE question at a time with A/B/C options, until the checklist is covered. Produce requirements.md (numbered FR/NFR) with explicit sign-off. No solutioning, no stories yet.
2. story-writer skill — Convert approved requirements.md into epics and INVEST user stories with Gherkin acceptance criteria and a traceability matrix. Produce stories.md.
3. story-reviewer skill — Independently review against INVEST, Definition of Ready, acceptance-criteria quality, and requirement coverage. Produce story-review.md.
4. jira-organizer skill — Turn stories.md into jira-plan.md: Initiative→Epic→ Story/Task hierarchy, Components from the codebase/integration wikis, label taxonomy, field mappings. Guidance only; no import file.

Helpers

The installer places wikikit.py (registry/sync/scaffold/status/commit/lint) and ingest.py (multi-format folder ingest) in $CODEX_HOME/tools (default ~/.codex/tools/); each skill also bundles them under its scripts/. Run as python3 ~/.codex/tools/wikikit.py … and python3 ~/.codex/tools/ingest.py <folder> --wiki <id>.

SharePoint / MCP sources

To let the Context Librarian read SharePoint (or other MCP sources), add an MCP server to your Codex config.toml (~/.codex/config.toml, or .codex/config.toml for a trusted project) — there is no single official SharePoint server, so point it at the Microsoft Graph / SharePoint MCP server you use and pass secrets via env, never inline:

[mcp_servers.sharepoint]
command = "npx"
args = ["-y", "<your-sharepoint-or-microsoft-graph-mcp-server>"]
env = { SHAREPOINT_TOKEN = "${SHAREPOINT_TOKEN}" }

If you have no MCP server, fall back to synced/exported files and ingest.py. DevLoop's requirements.md is intentionally richer than a plain spec (personas, FR/NFR with source traceability, scope, risks).