project-review

name

project-review

description

Skill for generating a standardised Wire Framework & Claude Code usage review for a client engagement. Activates when the user asks to review Wire usage, audit Claude Code activity, or generate a project review for a client. Gathers usage telemetry from BigQuery, Wire artifacts from the delivery GitHub repo, Jira tickets, and Fathom call transcripts, then produces a structured review document in wire/docs/reviews/.

Project Review Skill

On Activation

Before proceeding, append a one-line entry to .wire/execution_log.md:

| YYYY-MM-DD HH:MM | skill | project-review | activated | project review or audit work triggered this skill |

If .wire/execution_log.md does not exist, create it with the standard header first (see specs/utils/execution_log.md). If no .wire/ directory exists in the current repo, skip this step.

When This Skill Activates

Activate when the user asks to:

• Review Wire Framework usage for a client or project
• Audit Claude Code activity on an engagement
• Generate a project review, usage report, or Wire adoption report
• Understand how Wire was used (or not used) on a project

Keywords to watch for:

• "project review", "usage review", "Wire review", "Claude Code review"
• "how was Wire used on [client]", "audit [client] usage"
• "generate a review for [client]"

---

Step 1 — Gather Parameters

Before doing any research, collect the four required inputs. Ask all four in a single message:

To generate the project review I need four things:

1. Client name — e.g. "Client A" (used in the report title and repo clone path)
2. Wire delivery repo URL — the GitHub URL of the client's Wire delivery repo (e.g. https://github.com/rittmananalytics/rapha-delivery)
3. Project directory basename — the exact project_dir_basename value in BigQuery telemetry (e.g. rapha-dbt, halocollar-delivery). This is usually the name of the dbt or delivery repo folder on the consultant's machine.
4. Jira project key (optional) — the Jira project key for this engagement (e.g. RAP, HC). If you don't know it, leave blank and the skill will search Jira by client name.

Wait for all four answers before proceeding.

---

Step 2 — Gather Data (run all sources in parallel)

Once you have the four parameters, run the following research tasks in parallel. Do not wait for one before starting the others.

2a — BigQuery: Pull usage telemetry

Run this SQL, substituting {PROJECT_DIR_BASENAME} with the value from Step 1:

SELECT
    (FORMAT_TIMESTAMP('%F %T', coding_agent_prompts_fact.event_ts)) AS event_time,
    coding_agent_prompts_fact.prompt_sequence_in_session             AS prompt_seq,
    coding_agent_prompts_fact.slash_command_raw                      AS slash_command,
    coding_agent_prompts_fact.consultant_name                        AS consultant,
    coding_agent_prompts_fact.git_repo_canonical                     AS git_repo,
    coding_agent_prompts_fact.project_dir_basename                   AS project_dir,
    coding_agent_prompts_fact.truncated_display                      AS prompt_text,
    (CASE
        WHEN coding_agent_prompts_fact.slash_command_raw LIKE '/wire:%' THEN 'Yes'
        ELSE 'No'
    END) AS is_wire_command
FROM `ra-development.analytics.coding_agent_prompts_fact` AS coding_agent_prompts_fact
WHERE coding_agent_prompts_fact.project_dir_basename = '{PROJECT_DIR_BASENAME}'
GROUP BY 1, 2, 3, 4, 5, 6, 7, 8
ORDER BY 1

Use mcp__claude_ai_BigQuery_MCP__execute_sql_readonly with projectId: "ra-development".

The result set may be large. If it exceeds the context window, use a subagent to process the file:

Analyze the BigQuery result at <path>. It is JSON with schema:
{rows: [{event_time, prompt_seq, slash_command, consultant, git_repo, project_dir, prompt_text, is_wire_command}]}.

Produce:
1. Total prompts, distinct consultants, active days, date range
2. Wire command invocations — count per command, chronological list
3. Session-by-session narrative (group by date): date, consultant, wire commands used,
   key non-wire prompts (quote prompt_text verbatim for interesting ones)
4. Pre-wire prep patterns: what prompts immediately preceded Wire commands?
5. Post-wire correction patterns: what prompts immediately followed Wire commands?
6. Wire commands that should have been used but weren't (based on the prompt_text content)
7. Recurring non-wire patterns that appear 3+ times and have no Wire equivalent

2b — GitHub: Clone and explore the delivery repo

gh repo clone {REPO_URL} /tmp/{CLIENT_NAME_LOWER}-delivery-review

Then read the following (all in parallel):

• .wire/engagement/context.md — engagement overview, stakeholders, releases
• All .wire/releases/*/status.md files — release status and session history
• .wire/releases/*/brief.md files — release scope
• File tree of .wire/ to understand overall structure

2c — Jira: Pull all project issues

If a Jira project key was provided, query directly:

• Tool: mcp__claude_ai_Atlassian__searchJiraIssuesUsingJql
• cloudId: "https://rittmananalytics.atlassian.net"
• jql: "project = {JIRA_PROJECT_KEY} ORDER BY created ASC"
• maxResults: 100
• fields: ["summary", "status", "issuetype", "created", "assignee", "description"]

If no key was provided, or the key returns zero results, search by client name:

1. Call mcp__claude_ai_Atlassian__getVisibleJiraProjects to list all projects
2. Find the project whose name most closely matches the client name (case-insensitive substring match)
3. Use the discovered key in the JQL query above
4. If getVisibleJiraProjects is unavailable, try mcp__claude_ai_Atlassian__search with the client name as the query and filter results to Jira issues only

If the result is too large, use a subagent to extract: issue keys, types, statuses, assignees, and summaries grouped by epic/theme. Then map each issue to a Wire Framework phase (requirements, design, development, testing/QA, deployment, enablement) and identify phases with no Jira coverage.

2d — Fathom: Find relevant meetings

Use mcp__claude_ai_Fathom__search_meetings with the client name as the query and recorded_by: "[e-mail]". If no results, try with other team lead emails. Note: the delivery repo's .wire/engagement/calls/ directory contains pre-pulled Fathom transcripts — check there too.

---

Step 3 — Analyse the Data

Before writing the review, synthesise across all four sources:

Wire adoption rate:

• Total prompts vs. Wire command invocations → adoption percentage
• Which Wire commands were actually used and which were used correctly in sequence

Discovery phase comparison:

• What discovery artifacts exist in the repo vs. what Wire's canonical discovery flow would have produced
• Were /wire:problem-definition-generate, /wire:pitch-generate, /wire:release-brief-generate, /wire:sprint-plan-generate used? If not, what was done instead?

Release-by-release coverage:

• For each release in .wire/releases/, list which Wire generate/validate/review commands were invoked and which were skipped
• Identify development phases where Wire was bypassed but the work was clearly done (evidenced by BigQuery prompts or Jira tickets)

Pre/during/post-Wire prompt patterns:

• What context did consultants inject before or after Wire commands?
• Were Wire commands invoked correctly (generate before validate, validate before review)?
• Did consultants abort and restart Wire commands?

Gap analysis:

• Wire commands that exist and were directly applicable but never used
• Recurring free-form prompts that appear 3+ times and represent an unmet Wire need

---

Step 4 — Write the Review Document

Create the review at:

wire/docs/reviews/{YYYY-MM}-{CLIENT_NAME_LOWER}-wire-usage-review.md

Use today's year-month for {YYYY-MM} and the client name lowercased and hyphenated for {CLIENT_NAME_LOWER} (e.g. rapha, halo-collar).

The document must follow this structure exactly:

---

# Wire Framework & Claude Code Usage Review — {CLIENT_NAME}

**Engagement:** {CLIENT_NAME}
**Review Date:** {MONTH YEAR}
**Review Author:** Generated from BigQuery telemetry, {REPO_URL}, Jira {JIRA_PROJECT_KEY}
**Period Covered:** {START_DATE} – {END_DATE}
**Consultants:** {LIST}

---

## Contents

1. Executive Summary
2. Engagement Overview
3. Wire Framework Adoption — Quantitative Summary
4. Discovery Phase: Actual vs. Canonical Wire
5. Release-by-Release Wire Usage Analysis
6. Claude Code Prompt Patterns: Before, During and After Wire Commands
7. Gap Analysis: Wire Commands Never Used but Applicable
8. Recurring Manual Patterns — Candidates for New Wire Commands
9. Recommendations

---

## 1. Executive Summary

[3–5 paragraphs. Lead with adoption rate and the single most important finding.
Cover: what Wire was used for vs. what it should have been used for;
the highest-friction manual activities; top 3 recommendations.]

---

## 2. Engagement Overview

### Client & Scope
[1–2 paragraphs from context.md]

### Technology Stack
[Table: layer → technology, sourced from context.md]

### Delivery Releases
[Table: # | name | type | status | key scope — from context.md and release brief.md files]

### Key Stakeholders
[Table from context.md stakeholders section]

---

## 3. Wire Framework Adoption — Quantitative Summary

### Overall Statistics
[Table: total prompts, active days, consultants, Wire command invocations, Wire %, date range]

### Wire Commands Actually Used
[Table: command | count | context — list every invocation with the date and what preceded/followed it]

### Observations on Command Namespace / Version
[Note any deprecated command names used, version mismatches, or signs of confusion about command names]

---

## 4. Discovery Phase: Actual vs. Canonical Wire

### What Was Produced
[List all discovery artifacts found in the repo]

### Canonical Wire Discovery Flow vs. What Happened
[Table: wire artifact | wire command | what actually happened — one row per discovery artifact]

### Root Cause of Discovery Phase Gap
[Why weren't Wire discovery commands used? Timeline, adoption state, pre-Wire structure?]

### Specific Discovery Phase Gaps
[One paragraph per missing artifact type]

---

## 5. Release-by-Release Wire Usage Analysis

[One section per release. Each section covers:
- Wire commands used (count and list)
- Development approach (free-form summary)
- What worked / what didn't
- Wire status.md quality assessment
- Specific gaps for this release]

---

## 6. Claude Code Prompt Patterns: Before, During and After Wire Commands

### Pre-Wire Preparation Patterns
[What context did consultants prepare before invoking Wire commands?
Did they prep first or invoke then inject context?]

### During Wire Command Patterns
[How many corrective prompts followed each Wire invocation?
What types of corrections were needed?]

### Post-Wire Patterns
[What did consultants do immediately after Wire commands?
Recurring follow-up patterns?]

---

## 7. Gap Analysis: Wire Commands Never Used but Applicable

[Table: command | evidence of manual equivalent | estimated prompt saving
Only include commands where BigQuery evidence clearly shows manual equivalent work.
Include total estimated prompt saving at the bottom.
Then explain why each command wasn't used — based on evidence in the prompt sequence.]

---

## 8. Recurring Manual Patterns — Candidates for New Wire Commands

[One subsection per pattern. Each subsection:
- Pattern name and proposed command name
- Evidence: quote actual prompt_text examples, count occurrences
- Proposed command behaviour
- What input it needs and what output it produces]

Minimum 3 patterns. Only include patterns with 5+ occurrences.

---

## 9. Recommendations

[Numbered list R1–RN, each with:
- Title and priority (High / Medium / Low)
- Problem statement (what friction does this solve?)
- Proposed solution
- Implementation pointer (which spec file to create or modify)]

Group by: (a) new Wire commands, (b) improvements to existing commands,
(c) process/adoption changes.

---

## Appendix: Data Sources

[Table: source | content | records/size]
[Note the BigQuery query used, repo URL, Jira project key]

---

*Review generated {DATE}. Data cut-off: {BIGQUERY_MAX_DATE} for BigQuery telemetry.*

---

Step 5 — Commit and Push

After writing the review:

git add wire/docs/reviews/
git commit -m "docs: add {CLIENT_NAME} Wire Framework usage review ({YYYY-MM})"
git push origin main

Confirm the file path and commit SHA to the user.

---

Quality Checks Before Finishing

Before reporting done, verify:

• Every section of the template is populated (no [placeholder] text remaining)
• The Wire adoption rate is calculated correctly (Wire invocations ÷ total prompts)
• Every Wire command in the "used" table has at least one real date and context note
• The discovery phase table has one row per canonical Wire artifact
• Each release in the repo has its own section in Section 5
• Gap analysis only includes commands where BigQuery evidence is explicit
• Recurring patterns section only includes patterns with 5+ documented occurrences
• All recommendations are numbered, have a priority, and reference a spec file
• The review file is committed and pushed

---

Notes

• If BigQuery returns no rows for project_dir_basename, ask the user to confirm the exact value. Common variants: the basename of the client's dbt repo ({client}-dbt), the delivery repo name ({client}-delivery), or a project folder name. Run SELECT DISTINCT project_dir_basename FROM \ra-development.analytics.coding_agent_prompts_fact` WHERE project_dir_basename LIKE '%{client_fragment}%' LIMIT 20` to discover the right value.
• If the Jira project key is unknown or returns no results, search for the project by client name first: use mcp__claude_ai_Atlassian__search with the client name as the query, or use mcp__claude_ai_Atlassian__getVisibleJiraProjects to list all projects and match on name. Only skip Jira entirely if the MCP itself is unavailable — note that in the appendix.
• If Fathom returns no results, check .wire/engagement/calls/ in the delivery repo — Fathom transcripts are often pre-pulled there.
• The SQL has no person-name filter — it returns all consultants who worked in that project directory. The analysis should break down usage per consultant where multiple consultants appear.
• Review documents live in wire/docs/reviews/ in the Wire framework repo (not the client delivery repo).