Cadence Prep

Defaults

Prerequisites

The following credentials must be configured before use:

• Jira -- ATLASSIAN_EMAIL and ATLASSIAN_TOKEN in ~/.fe-skills/.env
• Slack -- SLACK_TOKEN and SLACK_COOKIE in ~/.fe-skills/.env

Verify Jira connectivity:

uv run --project .claude/skills/jira python .claude/skills/jira/scripts/issues.py list --max-results 1 --pretty

Not all sources are required. If Slack credentials are missing, build the agenda from Jira data alone -- note which sections are incomplete.

Workflow

Step 1: Parse Invocation

Parse the user's request to determine:

• Customer name (required): fuzzy-match against templates/customers.yaml keys. Check both the name field and jira_customer_variants for alternative spellings. If ambiguous, ask the user to confirm. If customer not found, follow the Unknown Customer procedure below.
• Cadence type (optional override): --weekly, --biweekly, --monthly, --qbr. If not specified, use the customer's cadence.type from customers.yaml.
• Date (optional): defaults to today. Used for the document filename and header.
• Dry-run flag: if --dry-run is specified, preview the agenda in chat without saving to disk or publishing to Confluence.
• HTML flag: if --html is specified, also generate a styled HTML version for screen-sharing (in addition to Confluence + local markdown).

Read the customer entry from templates/customers.yaml to get: jira_customer, slack_channels, cadence schedule, contacts.

Compute the lookback window from cadence type:

Calculate the Unix timestamp for the --oldest parameter: current_time - (lookback_days * 86400).

Step 2: Gather Jira Data

Run these data-gathering operations:

Open issues:

uv run --project .claude/skills/jira python .claude/skills/jira/scripts/issues.py list \
  --customer "<jira_customer>" --with-comments --max-results 200 --pretty

Recently closed (within lookback window or last 30 days, whichever is greater):

uv run --project .claude/skills/jira python .claude/skills/jira/scripts/issues.py search \
  --jql "project = WB AND status IN (Done, Closed, Resolved) AND updated >= -30d ORDER BY updated DESC" \
  --customer "<jira_customer>" --max-results 50 --pretty

FE-UPDATE status for each open issue:

uv run --project .claude/skills/jira python .claude/skills/jira/scripts/issues.py fe-updates \
  --key WB-XXX --pretty

From the gathered data, compute:

• Open issue table: key, summary, type, priority, status, last activity date, latest FE-UPDATE status
• Recently resolved (within cadence window): issues that moved to Done/Closed/Resolved/Merged since the last cadence. These are wins to highlight — for each one, write a brief customer-facing summary of what was fixed or delivered and why it matters to them. These feed into the "What's New" section as good news to share.
• Recently closed (broader, last 30 days): key, summary, resolved date — for the support tickets summary table
• Stale issues: open issues where the most recent comment (excluding SE's own FE-UPDATE comments) is older than 30 days
• Issue URLs: https://coreweave.atlassian.net/browse/WB-XXX

Step 2.5: Gather Asana Actions and RAID Items

This step gathers two data sources from Asana to feed into the meeting agenda. Both are optional -- cadence-prep works without Asana data.

2.5a: Asana Actions (from customer project)

uv run --project .claude/skills/asana python .claude/skills/asana/scripts/query.py tasks \
  --project-gid <action_tracker_id> --limit 100 --pretty

Process the response:

• Open actions (incomplete tasks in To Do, In Progress, Waiting on Customer, Waiting on Eng, Scheduled/Future): feed into the Action Items table in the agenda.
• Recently completed (tasks with completed_at within the cadence lookback window): feed into "What's New" section as wins to highlight.
• Format each action for the Action Items table: | # | Action | Owner | Due | Status | Source |
• Map Asana sections to Status: To Do -> "Open", In Progress -> "In Progress", Waiting on Customer -> "Waiting (Customer)", Waiting on Eng -> "Waiting (Eng)", Scheduled/Future -> "Scheduled"

If action_tracker_id is "PLACEHOLDER" or not present: skip Asana actions, note "Asana actions not configured -- add action_tracker_id to templates/customers.yaml" in the Action Items section.

2.5b: Open RAID Items (from RAID project)

uv run --project .claude/skills/asana python .claude/skills/asana/scripts/query.py tasks \
  --project-gid <raid_tracker_id> --limit 100 --pretty

Process the response:

• Filter to incomplete tasks only.
• Group by section (Risks, Assumptions, Issues, Dependencies).
• Risks and Dependencies with Status = "Open" or "Accepted": feed into the RAID Log table in the agenda. These are the items that should be discussed in the meeting.
• Include custom field values: Impact, Likelihood, Status.
• Format for RAID Log table: | Type | Item | Owner | Status |

If raid_tracker_id is "PLACEHOLDER" or not present: skip RAID items, note "RAID log not configured -- add raid_tracker_id to templates/customers.yaml" in the RAID Log section. This is graceful degradation -- cadence-prep works without RAID.

Step 3: Gather Slack Data

Read channel history for each of the customer's Slack channels (from customers.yaml slack_channels):

uv run --project .claude/skills/slack python .claude/skills/slack/scripts/channels.py history \
  --channel <channel_id> --limit 50 --oldest <lookback_timestamp> --pretty

For threads with reply_count > 0, fetch the full thread to get context:

uv run --project .claude/skills/slack python .claude/skills/slack/scripts/threads.py replies \
  --channel <channel_id> --ts <thread_ts> --limit 50 --pretty

From the gathered data, produce:

• What's New summary: group notable threads by topic, summarize each in 1-2 sentences
• Slack permalinks: construct from channel_id + message timestamp: https://coreweave.slack.com/archives/{channel_id}/p{ts_without_dot}
• Filter out low-signal messages (bot messages, join/leave notifications, reactions-only)

Resolve Slack identities: Slack messages contain user IDs in the format <@UXXXXXXXX>. Before writing any output, resolve these to real names using the Slack skill:

uv run --project .claude/skills/slack python .claude/skills/slack/scripts/users.py info --user UXXXXXXXX --pretty

Cache resolved names for the session to avoid repeated lookups. Replace all <@UXXXX> references with the person's display name in the final output.

Handling edge cases:

• If a channel ID is PLACEHOLDER, skip it and note: "Slack channel ID for [channel name] is a placeholder. Update templates/customers.yaml with the real channel ID."
• If no Slack channels are configured or all return empty, note: "No Slack activity found in the lookback window."
• If Slack credentials fail, note which sections are incomplete and continue with Jira data.
• If a user ID cannot be resolved, leave it as <@UXXXX> rather than guessing.

Step 4: Carry Forward from Previous Docs

Look for existing cadence docs:

ls -1 customers/<customer>/calls/*.md 2>/dev/null | sort | tail -1

If a previous doc exists, read it and extract:

• Attendees table: carry forward as-is (SE edits at presentation time)
• Action Items: items with Status = "Open" or "In Progress". Cross-reference with current Jira data: if an action item references a WB-XXX key that is now in Done/Closed/Resolved status, mark it "Resolved" in the new doc. Items without Jira links carry forward unchanged.
• RAID Log: If RAID project is configured (Step 2.5b returned data), use live RAID data from Asana instead of carrying forward from previous doc -- the Asana RAID project is the source of truth. If RAID project is NOT configured, carry forward items with Status != "Closed" or "Resolved" from the previous doc (backward compatible).
• Key Initiatives: carry forward entire section as-is

If no previous doc exists (first run for this customer), note: "First cadence doc for [Customer]. No carry-forward data."

Resolution heuristics:

• An action item is "resolved" if its linked Jira issue is in Done/Closed/Resolved status
• An action item without a Jira link carries forward unless marked "Done" in the previous doc
• If parsing the previous doc fails (user may have manually edited it), carry forward nothing rather than corrupt data -- note "Could not parse previous doc -- starting fresh for carried items."

Step 5: Gather Product Updates

Gather product updates from the lookback window, then filter by the customer's deployment_type from customers.yaml.

Source 1: W&B Release Notes (docs.wandb.ai)

Fetch SDK and server release notes via WebFetch: