rice-score

name	rice-score
description	Research a RHAISTRAT Feature ticket and recommend RICE scores (Reach, Impact, Confidence, Effort). Read-only — outputs structured recommendations for the Python orchestrator to write to Jira. Used headlessly by quality_gate.py.

RICE Scoring for Release Quality Gate

Research a single RHAISTRAT Feature ticket and recommend RICE scores. This skill is read-only — it does NOT write to Jira. The Python orchestrator handles all Jira writes.

Reference Files

Read these before scoring:

• references/rice-rubric.md — Full RICE rubric with scoring scales
• references/jira-fields.md — Jira custom field IDs and API reference
• references/calibration-examples.md — Scored examples from our batch for anchoring

Input

A single Jira key passed as the skill argument: /rice-score RHAISTRAT-1745

Workflow

Step 1: Fetch the Target Ticket

Fetch the full ticket with all relevant fields:

curl -s -u "$JIRA_USER:$JIRA_TOKEN" \
  "$JIRA_SERVER/rest/api/3/issue/{{TICKET_KEY}}?fields=summary,description,status,comment,issuelinks,attachment,issuetype,parent,priority,labels,customfield_10862,customfield_10836,customfield_10838,customfield_10637,customfield_10864"

Step 2: Build Calibration Data

Fetch existing RICE-scored issues for relative positioning:

curl -s -u "$JIRA_USER:$JIRA_TOKEN" "$JIRA_SERVER/rest/api/3/search/jql" \
  -H "Content-Type: application/json" \
  -d '{"jql": "project = RHAISTRAT AND labels = \"strat-creator-human-sign-off\" AND cf[10862] IS NOT EMPTY ORDER BY key ASC", "fields": ["summary","customfield_10862","customfield_10836","customfield_10838","customfield_10637","customfield_10864"], "maxResults": 20}'

Format as calibration anchors:

- RHAISTRAT-XXXX (Summary): R=X, I=X, C=XX%, E=X, RICE=X.X

Also read the examples in references/calibration-examples.md for additional context.

Step 3: Research the Ticket

Strategy and review attachments are the richest source of evidence. The Jira description has limited formatting (ADF), but the attached strategy and review documents contain the full analysis from the strat-creator pipeline.

3a. Download and read attachments (highest priority)

List attachments from the ticket's attachment field. Look for:

• Strategy document (*-strategy.md or the main .md attachment): Contains the full refined strategy — HOW, dependencies, impacted teams/components, effort estimates, risks, acceptance criteria, non-functional requirements, and scope boundaries.
• Review document (*-review.md): Contains 4-dimension review scores (feasibility, testability, scope, architecture, each 0-2) with detailed prose feedback from independent reviewers. Total score out of 8.

Download each attachment:

curl -s -u "$JIRA_USER:$JIRA_TOKEN" \
  "$JIRA_SERVER/rest/api/3/attachment/content/{attachmentId}" -L

How attachment content maps to RICE dimensions:

• Strategy's effort estimate and team count → RICE Effort
• Strategy's dependencies and risks → RICE Effort and Confidence
• Strategy's scope and acceptance criteria → RICE Reach (who benefits)
• Review's total score (X/8) → RICE Confidence (7-8/8 = 75-100%, 3-6/8 = 50-75%)
• Review's feasibility verdict → RICE Effort (reject/revise = bump effort up)
• Review's scope verdict → RICE Reach and Effort (scope too large = higher effort)

3b. Read the description

The Jira description may contain a summary of the strategy or the business need. It complements but does not replace the attachment content.

3c. Follow issuelinks to linked RFEs

Read linked RHAIRFE tickets for customer context, approval status, and priority labels. RFE comments often contain customer evidence that informs Reach and Impact.

3d. Read all comments

Check for reviewer feedback, strategy review scores, sizing disagreements, and any RICE scoring discussions or prior justifications.

Step 4: Apply the RICE Rubric

Score each dimension using references/rice-rubric.md. Key principles:

• Calibrate, don't score in isolation — justify relative to sibling tickets
• Trust strategy reviews over original estimates — reviews reflect multi-reviewer assessment
• Confidence reflects evidence quality, not feature importance
• Effort captures coordination cost — team count and dependency chains matter
• Flag re-scoring triggers — note what would change the score

Step 5: Output Structured Recommendation

Output the recommendation in this exact format so the Python orchestrator can parse it:

RICE_RECOMMENDATION_START
TICKET: {{TICKET_KEY}}
REACH: <value>
IMPACT: <value>
CONFIDENCE: <value as percentage: 50, 75, or 100>
EFFORT: <value>
EXPECTED_RICE: <calculated score>
JUSTIFICATION:
<Multi-line justification text explaining each dimension's score,
calibration context vs siblings, key evidence, and re-scoring triggers.>
RICE_RECOMMENDATION_END

Valid values:

• Reach: 1, 3, 5, 8, 13
• Impact: 1, 3, 5, 8, 13
• Confidence: 50, 75, 100
• Effort: 1, 2, 3, 5, 8, 13

Error Handling

• Ticket not found: Output RICE_ERROR: Ticket {{KEY}} not found
• RICE fields already set: Output RICE_ALREADY_SCORED: {{KEY}} R=X I=X C=X% E=X RICE=X.X
• No attachments: Score from description, links, and comments. Note reduced confidence.
• Effort = 13: Flag in justification that the feature must be broken down.