Documentation Migration Map
Inventory and classification for conversational documentation
Conversational Documentation Migration Map
This inventory records the baseline for issue #1937. It is intentionally classification-based: command blocks are moved or rewritten according to their reader and task, not deleted by raw count.
| Surface | Baseline classification | Destination/action |
|---|---|---|
| `docs/agents/CLI_USAGE.md` | Agent/operator reference | Moved to `docs/agents/CLI_USAGE.md` |
| `docs/agents/cli-reference.md` | Agent/operator reference | Moved to `docs/agents/cli-reference.md` |
| Discovery and kernel procedures | Agent-internal | Indexed agent reference; public pages describe the conversational result |
| Welcome/homepage | End-user orientation | Lead with conversation, expected behavior, approval, and result |
| Start-here and quickstart | End-user tutorial | Lead with asks; retain bootstrap escape hatch |
| Provider handoff | End-user how-to plus operator details | Conversational primary path; operational flags in agent corpus |
| Onboarding validation | End-user verification | Agent runs probe and translates evidence |
| Framework entry points | End-user how-to | Ask/outcome/verification pattern |
| Install pages | User-required direct touch | Retain justified bootstrap commands |
| Troubleshooting | Recovery escape hatch | Commands allowed when clearly labeled |
| Development/contributing | Maintainer | Retain commands with explicit audience |
| Blog and release notes | Historical | Preserve wording and command history |
| Tests and fixtures | Example fixture | Excluded from prose migration |
Audit baseline
At issue intake, `docs/_manifest.json` exposed both CLI references and `docs/config.json` used three command-first checklist items plus a CLI quick link. Core onboarding pages directly instructed users to run wizard, probe, discover, and show operations. The post-migration audit is generated by `npm run docs:audience:audit`.
The generated audit enumerates every path under `docs/getting-started/` and every Markdown filename containing `quickstart`, `how-to`, or `howto`. Each is classified as the canonical onboarding page, linked to canonical onboarding, contributor/maintainer, historical, or specialized-needs-review. The audit also records whether the page contains the preferred `use all --provider` and `aiwg-regenerate` hookup signals; this makes remaining migrations explicit without treating every specialized tutorial as a duplicate install guide.
The inventory also records every Markdown page containing an `aiwg` command, the number of direct-touch mentions, and the number of agent/operator mentions. During public-source staging, any non-historical public page that retains a command outside the direct-touch allowlist receives a visible execution notice: those commands are agent/operator detail, and conversational users should state the outcome they want. This makes retained exceptions explicit instead of relying on readers to infer the audience. Historical release notes, contributor documentation, and the source files themselves are not mechanically rewritten.
Before and after
The audit baseline is stored in `docs/public-docs.json` and was calculated from the repository HEAD at issue intake. The same audit reports current values:
| Signal | Before | Current target |
|---|---|---|
| Command mentions across the seven core public journeys | 51 | substantially reduced; retained commands are bootstrap or recovery |
| Command-first homepage checklist items | 3 | 0 |
| CLI/agent-reference entries in public navigation | 2 | 0 |
Operational coverage is retained in `docs/agents/`, the npm `docs/` payload, and the signed release pipeline's `reference` bundle.