Fill Cycle — Velocity Pipeline

Load the current Linear cycle, compute the capacity gap against proven velocity, scan 4 backlog sources, and propose a ranked fill plan.

Target: $ARGUMENTS

Linear Label IDs (cached)

Step 0: Determine Mode

Step 1: Fetch Current Cycle from Linear

linear.list_cycles(teamId: "<LINEAR_TEAM_ID>")

Identify the current cycle (type "current" or date range containing today). Extract: cycle_id, name, start, end, scopeTotal, completedTotal.

Then fetch all issues in the cycle:

linear.list_issues(
  team: "<LINEAR_TEAM_ID>",
  cycle: "<cycle_id>",
  first: 50
)

Compute:

• cycle_days = (end - start).days
• days_remaining = (end - today).days
• pts_committed = sum of all issue estimates in cycle
• pts_done = sum of estimates for Done/Canceled issues

Step 2: Read Velocity Baseline

Read .claude/state/velocity.json.

Extract:

• rolling_avg_pts_per_day = rolling_avg_3.pts_per_day
• target_utilization (default 0.80)

Compute capacity gap:

capacity = rolling_avg_pts_per_day × cycle_days
target = capacity × target_utilization
gap = target - pts_committed

If gap <= 0: report "Cycle is fully loaded" and stop (unless --check which still reports).

Step 3: Scan Backlog Sources

Scan 4 sources for candidate items. Each source contributes candidates with metadata.

Source 1: Linear Backlog (unassigned, no cycle)

linear.list_issues(
  team: "<LINEAR_TEAM_ID>",
  first: 50
)

Filter for issues where:

• cycle is null (not in any cycle)
• status is "Backlog" or "Todo"
• assignee is null or matches current user

Extract: id, identifier, title, priority, estimate, labels.

Source 2: BACKLOG.md (post-MVP items)

Read BACKLOG.md from repo root.

Parse all CAB-XXXX entries from tables and lists. Extract:

• Ticket ID, title, section (Post-MVP, Vision, MCP, Community, etc.)
• Map section to roadmap theme:
  • "MCP Gateway" / "Features Avancées" → gateway
  • "Community" → community
  • "Observability" → observability
  • "Portal & UX" → platform
  • "Infra & DevOps" → platform
  • "Architecture" → platform
  • "Vision & Strategy" → platform

Source 3: plan.md (parked backlog sections)

Read plan.md from repo root.

Parse entries from:

• "Backlog — Long-term" section (current cycle)
• "Backlog" section of next cycle
• Any item without a checkbox (parked, not committed)

Extract: ticket ID, title, estimate (if present), priority.

Source 4: CAPACITY-PLANNING.md (phased roadmap)

Read docs/CAPACITY-PLANNING.md.

Parse unchecked [ ] items from remaining phases:

• Phase 2.6, 4, 5, 6, 7, 8, 9, 10, 11, 14
• Extract: ticket ID, title, phase name

Map phase to roadmap theme:

• Phase 2.6 (Cilium), 5 (Multi-env) → platform
• Phase 4 (OpenSearch) → observability
• Phase 7 (Security) → gateway
• Phase 8 (Portal) → dx
• Phase 9 (Ticketing) → platform
• Phase 6 (Demo) → community
• Phase 10, 11 (Resource Lifecycle) → platform
• Phase 14 (GTM) → community

Source 5 (conditional): Notion — only with --theme

If --theme <name> is provided:

notion.notion-search(query: "STOA roadmap <theme>")

Extract any actionable items not already in Linear. Flag as "Notion-sourced" in output.

Step 4: Deduplicate & Score Candidates

Merge all candidates. Deduplicate by CAB-XXXX identifier (Linear source wins if conflict).

Score each candidate (higher = better fit):

Sort by score descending.

Step 5: Classify into 3 Buckets

Bucket 1: Quick Wins (promote now)

• Estimate <= 8 pts
• Already in Linear (has issue ID)
• Not an EPIC or MEGA (no [MEGA] or [EPIC] in title)

Bucket 2: MEGAs (need council + decompose)

• Estimate > 13 pts OR title contains [MEGA] / [EPIC]
• Suggest: run /council CAB-XXXX then /decompose CAB-XXXX

Bucket 3: Unpointed (flag for estimation)

• No estimate value
• Action: "Estimate before promoting"

Step 6: Build Fill Plan

Cap at top 15 candidates total across all buckets.

Within the gap budget:

1. Fill with Quick Wins first (sum estimates, stop when gap reached)
2. If gap remains > 13 pts: suggest 1 MEGA
3. List remaining Unpointed items as "estimate to unlock"

Group MEGAs by roadmap theme for readability.

Step 7: Execute (mode-dependent)

Interactive mode (default)

Display the fill plan. For each Quick Win, ask: "Promote Y/N?" On Y: linear.update_issue(id, cycleId: "<current_cycle_id>")

Auto mode (--auto)

Promote all Quick Wins automatically:

linear.update_issue(id: "<issue_id>", cycleId: "<current_cycle_id>")

Log each promotion in operations.log.

Check mode (--check)

Display the capacity report only. No Linear writes. Exit.

Themed mode (--theme <name>)

Same as interactive but filtered by theme. Includes Notion results if available.

Step 8: Update State Files

After any promotions:

1. Run /sync-plan to refresh plan.md with newly promoted issues
2. Append to operations.log:

   STEP-DONE | step=fill-cycle task=velocity-pipeline promoted=N gap_before=X gap_after=Y

Step 9: Report

Capacity Report — Cycle N (dates)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Velocity baseline (rolling 3-cycle avg):
  Avg: XX.X pts/day | Target utilization: 80%

Current cycle:
  Committed: XXX pts | Done: YYY pts (ZZ%)
  Capacity: AAA pts | Target: BBB pts (80%)
  Gap: CCC pts (DD% under-loaded)

Fill Plan:
━━━━━━━━━
Quick Wins (promote now):                    Σ XX pts
  1. CAB-XXXX: Title (N pts, P2) ............. [score: 14]
  2. CAB-YYYY: Title (N pts, P1) ............. [score: 17]

MEGAs (→ /council + /decompose):             Σ XX pts
  [gateway] CAB-ZZZZ: Title (21 pts, P2)
  [platform] CAB-WWWW: Title (34 pts, P3)

Unpointed (estimate to unlock):
  CAB-VVVV: Title (no estimate, P2)

Summary:
  Promotable: XX pts | After fill: YYY/ZZZ pts (WW%)
  MEGAs needing council: N
  Unpointed needing estimate: M

MCP Budget

Rules

• velocity.json is the single source of truth for capacity baseline
• 80% target utilization leaves 20% for unplanned work (configurable in velocity.json)
• Cap proposals at 15 — avoid overwhelming fill plans
• MCP budget: 3-4 read calls per run — batch, don't poll
• BACKLOG.md items stay unpointed until promoted (estimation happens at promotion time)
• No auto-promote in --check mode — always read-only (CI-safe)
• No Notion call without --theme — avoid unnecessary MCP latency
• Quick Wins first — fill with small items before suggesting MEGAs
• Never promote without estimate — Unpointed items must be estimated first
• Log all promotions in operations.log for audit trail