Release Planning Analysis

---

name: release-planning-analysis description: Write actionable recommendations for a release planning risk assessment — reads pre-computed check results and produces narrative per person and per feature allowed-tools: Read, Write user-invocable: false

release-planning: Analysis

Purpose

Read the pre-computed planning risk checks from checks.json and write actionable, natural-language recommendations. All arithmetic (capacity, timeline, composite scoring) has already been computed by run-checks.py — this agent only interprets the results and writes narrative.

When to Spawn

The parent release-planning skill spawns this agent during Phase 5b, after run-checks.py has produced {WORKDIR}/checks.json.

Parameters

Substituted by the parent before spawning:

Placeholder	Description
{WORKDIR}	Work directory path
{VERSION}	OCP release version (e.g., 5.0)

Instructions

Step 1: Read Data

Read {WORKDIR}/checks.json. This file contains all computed check results:

• meta — summary counts (total features, risk levels, overall risk)
• data_quality — per-feature gate status (PASS/WARN/FAIL)
• capacity — per-person assigned SP vs remaining capacity
• timeline — per-feature remaining work vs time left
• assignment — SPOFs and unassigned work
• bug_load — unassigned Blocker/Critical bugs
• sizing — T-shirt size mismatches
• composite — per-feature composite risk level
• unknown_contributors — people with work assigned who aren't in the team roster

Step 2: Write Recommendations

Write {WORKDIR}/recommendations.json with this structure:

{
  "executive_summary": "<3-5 sentences covering: overall risk verdict, top 2-3 risks, most critical action>",
  "per_person": [
    "<actionable recommendation per overloaded or SPOF person, referencing their SP numbers and features>"
  ],
  "per_feature": [
    "<actionable recommendation per at-risk feature, starting with HIGH composite risk>"
  ],
  "team_level": [
    "<overall team actions — capacity rebalancing, scope cuts, process changes>"
  ]
}

Writing Guidelines

• Be specific: Reference actual numbers from checks.json. "Alice has 25 SP assigned but only 16 SP capacity" not "Alice is overloaded."
• Be actionable: Every recommendation should name a person, a feature, and a concrete action. "Move 9 SP from Alice to Bob" not "Rebalance load."
• Prioritize HIGH composite risk features: Address them first in per-feature recommendations.
• Flag unknown contributors: If unknown_contributors is non-empty, note that these people have work assigned but aren't in the team roster — their velocity assumptions may be wrong.
• Data quality failures: For features with FAIL status, recommend specific next steps (create stories, point existing stories) rather than just noting the failure.
• Don't repeat numbers the report already shows: The tables will have the data. Recommendations should interpret and advise, not restate.
• Use bare Jira keys only: Write OCPSTRAT-2607 not [OCPSTRAT-2607](url). The report assembly script adds Jira links automatically — pre-linking causes broken nested links.

Important Notes

• This agent does NOT compute any numbers — all arithmetic is in checks.json
• This agent does NOT build hierarchies or read Jira data files
• This agent does NOT write markdown sections or sentinel blocks
• Output is a single JSON file with four keys
• Use natural language — be conversational and actionable