team-update

Post project updates to team chat, gather feedback, triage responses, and plan next steps. Adapts to available tools (chat, git, issues, tasks). First run discovers tools and saves a playbook; subsequent runs execute from the playbook. Trigger with 'team update', 'post update', 'sync with team', 'st

By jezweb · 962 installs

npx skills add jezweb/claude-skills --skill team-update

Source repository · Upstream listing

Team Update Post project updates to team chat, read and triage feedback, and plan next steps. Adapts to whatever communication and project tools are available. The Playbook Pattern This skill uses a playbook file at .claude/team update playbook.md in the project root. First run : Discover available tools, ask the user for preferences, execute, save a playbook Subsequent runs : Read the playbook, skip discovery, execute directly User edits playbook : Changes take effect immediately (channel, style, triage rules) The playbook is plain markdown the user can edit. It captures tool configuration, channel preferences, message style, triage rules, and last run metadata. Phase 0: Playbook Check Check if .claude/team update playbook.md exists in the project root. If it exists : Read it. All tool configuration, channel info, message style, and triage rules are in there. Jump to Phase 1 with known config. If it does not exist : Run Phase 0b (Discovery) first. Phase 0b: Discovery (First Run Only) Detect available capabilities by checking what MCP tools are connected and what local tools exist. Read [references/discovery patterns.md](references/discovery patterns.md) for detection patterns. Capability Detection Check for each capability category: Capability How to detect Fallback if missing Chat MCP tools matching chat , slack , discord , teams Output formatted text for manual posting Git git rev parse is inside work tree Skip commit summaries Issues MCP tools matching github , linear , jira Skip or output text list Tasks MCP tools matching tasks , todos , asana Skip or output text list Knowledge Basalt Cortex ( ~/Documents/basalt cortex/ ) Search for related knowledge notes User Preferences (Ask) After discovery, ask the user to confirm: 1. Channel : Which chat space/channel to post updates to? (List discovered options) 2. Identity : Post as yourself or a bot identity? (e.g. Anthro Morphic) 3. Repo scope : Which repo(s) to track? (Default: current repo) 4. Message style : Brief standup, detailed changelog, or conversational update? 5. Team context : Any team members to be aware of? (Names, roles) 6. Triage rules : Any custom rules for handling feedback? (Or use defaults) Save Playbook After discovery + user preferences, save the playbook to .claude/team update playbook.md . Read [references/playbook format.md](references/playbook format.md) for the full template. Fill in all sections with discovered tools and user preferences. If a capability was not found, note it as "Not available" in the playbook. Ensure .claude/ is in .gitignore (the playbook contains space IDs and preferences that shouldn't be committed). Phase 1: Gather Context Collect information from all available sources. Use the playbook's "Data Sources" section to know which tools to call. 1a. Determine Time Window Check the playbook's "Last Run" section for the last update timestamp. If no previous run, ask the user how far back to go (default: 24 hours). 1b. Collect Data Gather from each available source: Source What to collect Tool Git commits git log oneline since=<last update Bash Chat messages Messages in the configured channel since last update Chat MCP tool from playbook Open issues/PRs Recently updated or newly created items Issue tracker from playbook Active tasks In progress or recently completed tasks Task tracker from playbook 1c. Summarise for Drafting Group commits by theme (not individual hashes). Note any issues closed, PRs merged, or tasks completed. Flag any team messages that need response. Phase 2: Draft and Post Update 2a. Compose Message Read [references/message composition.md](references/message composition.md) for message patterns. Follow the playbook's "Message Style" section. General principles: Lead with what shipped or changed, not process Group related commits into themes Mention specific people only when relevant Keep it scannable (short paragraphs, bullet points) Write like a teammate, not an AI — no emoji spam, no hedging, no over explaining context the team already has. See the Communication Style section in [references/message composition.md](references/message composition.md). 2b. Preview and Approval Show the draft to the user. Include: The composed message Which channel it will be posted to Whether it's a new message or thread reply APPROVAL REQUIRED : Never post to external channels without explicit user approval. 2c. Post Post the approved message using the chat tool from the playbook. If no chat tool is available, output the formatted message for the user to post manually. 2d. Update Playbook After posting, update the "Last Run" section of the playbook with: Current timestamp Thread key (if applicable) Brief summary of what was posted Phase 3: Read and Triage Feedback 3a. Read Responses Read messages from the configured channel since the update was posted. Look for: Direct replies to the update New messages in the channel that reference the project Reactions or acknowledgements 3b. Classify Feedback Read [references/feedback triage.md](references/feedback triage.md) for classification patterns. Use the playbook's "Triage Rules" section. Default classification: Type Action Bug report Create issue (label: bug) Feature request Create issue (label: enhancement) Question Draft reply for user approval Blocker Flag immediately, suggest resolution Acknowledgement Note, no action needed Off topic Ignore 3c. Present and Act Present a summary of feedback with proposed actions: APPROVAL REQUIRED : Never create issues, post replies, or take external actions without explicit user approval. Phase 4: Plan Next Steps 4a. Prioritise From all gathered context (commits, issues, tasks, feedback), identify: 1. Blockers : Things preventing progress 2. High priority : Bugs, urgent requests, deadlines 3. Next up : Planned work, feature requests 4. Backlog : Nice to haves, low priority items 4b. Present (Not Posted) Show the prioritised list to the user. This is for the user's planning only do NOT post it to any channel. Autonomy Rules Action Level Read/create playbook Just do it Discover tools (first run) Just do it, confirm preferences with user Gather context (git, chat, issues) Just do it Draft update message Just do it, show preview Post to external channel Approval required Create issues/tasks Approval required Post replies to team Approval required Update playbook "Last Run" Just do it Present plan/priorities Just do it Graceful Degradation The skill works at any level of tool availability: Available tools Experience Git only Summarise commits, output text for manual posting Git + Chat Full post/read cycle, manual issue creation Git + Chat + Issues Full cycle with automated issue creation Git + Chat + Issues + Tasks + Knowledge Full cycle with persistent tracking Operating Modes The skill responds to different trigger phrases: Trigger Behaviour "team update" / "post update" Full cycle: gather, draft, post, read feedback "check team chat" / "what did the team say" Phase 3 only: read and triage feedback "standup" / "sync with team" Phase 1 + 2: gather and post, skip feedback "project update" Phase 1 + 2 + 4: gather, post, plan (skip feedback triage) "feedback loop" Phase 3 + 4: read feedback and plan Reference Files When Read First run — saving playbook [references/playbook format.md](references/playbook format.md) First run — detecting tools [references/discovery patterns.md](references/discovery patterns.md) Composing update messages [references/message composition.md](references/message composition.md) Classifying team feedback [references/feedback triage.md](references/feedback triage.md)