[hsuchifeng]

Summary

Before: several guide pages described behavior the CLI no longer has. The biggest one: getting started said init pulls, so a new member following it saw teamai doctor fail ("Skills delivered to claude ✖"). The manifest/projects.yaml examples also lacked version: 1, so pasting them broke the next pull.

After: every fixed statement matches the code, and the join flow (init → pull → doctor → list) passes as written. English and zh-CN pages get the same fixes.

Fixes:

• init no longer pulls ([bug] init without --agent in a non-interactive shell reports agents synced but delivers nothing into the project Tencent/teamai-cli#1065): getting started and the member guide run teamai pull before doctor.
• manifest/projects.yaml examples (sharing, knowledge) gain the required version: 1 (src/projects.ts).
• Recall: teamai recall enable/disable set the local recallEnabled override. TEAMAI_RECALL_DISABLED=1 only mutes hooks, and pull still installs recall. The --update-quality draft is <file>.md.draft.md.
• Distribution: the init . layout (directories, not env.yaml / mcp.yaml), what push sends (skills, rules, agents, env), that roles can declare env/hooks/mcp/models/docs/wiki too, and that a tag subscription filters rules instead of adding them.
• coAuthorEnabled lives in the scope's local config. The dashboard's Team Context has its own data source. The FAQ quotes push's real message. The docs.localDir default depends on scope. The ack body lists every field. Init records the provider only when it creates teamai.yaml.

Type of Change

• Documentation only

Test Plan

Docs only, so no e2e record is required. Each fix was checked against the source, and the main examples were run with the built CLI in an isolated HOME, against a local bare team repo through --provider git:

• teamai init <url> --provider git --agent claude, then teamai doctor: this failed "Skills delivered to claude" before the pull. After teamai pull it printed "All checks passed!", and teamai list skills --source repo showed the skill.
• teamai push --skill ~/.claude/skills/hello filled in the frontmatter and pushed the branch.
• With the projects snippet as written, pull failed with Invalid projects manifest: version: Required. With version: 1 it passed.
• tags subscribe x, then pull, printed "Synced 1 rule(s) (skipped 1 by tags)".

Notes for Reviewers

Two crashes came up during testing and are left out of this PR. Each prints a raw stack trace instead of an error message: teamai status / teamai list before init (an uncaught NotInitializedError), and teamai packages without a package.json.

---

Generated by Claude Code

[github-actions]

Code Erosion Report

Informational — never blocks the merge. scb-check==0.2.0 SlopCodeBench metrics; method & caveats in docs/dev/ci-code-erosion.md.

Metric	Value	Human band	Agent band	Reading
Verbosity*	0.065	0.15	0.33	human
Erosion	0.703	0.31	0.68	agent
Cognitive erosion	0.875	–	–	–

Scanned 319 files / 69098 SLOC · high-CC functions 516/5692

* On TypeScript, verbosity is partial and the bands are Python-calibrated.

Rule hits (TS verbosity layer)

Standalone ast-grep, separate from the number above.

Rule	Hits
unnecessary-else-after-return	38
empty-catch-block	5

43 distinct lines flagged across src/.

[hsuchifeng]

Codex PR Review is red, but the failure is not from this PR. The job dies in openai/codex-action setup ("Failed to read server info from /home/runner/.codex/.json") before any review runs. That points to the Codex review credentials/config not being available on this fork. This PR changes only docs, so no fix belongs here. A re-run would hit the same missing setup.

---

Generated by Claude Code