confluence-docs-review

name

confluence-docs-review

description

Manually triggered skill that browses Confluence documentation using Atlassian MCP tools, extracts coding practices and technical knowledge, classifies them for playbook relevance, and auto-invokes packmind-update-playbook with structured findings. Use when wanting to mine Confluence pages for conventions, architectural decisions, best practices, or technical knowledge that should be captured in the Packmind playbook. Supports non-interactive mode via `CI=true` or `--non-interactive` for scheduled runs.

Confluence Documentation Review

Browse Confluence documentation to extract coding practices, technical knowledge, and conventions, classify them for playbook relevance, and feed actionable findings into the packmind-update-playbook workflow.

Workflow: Phase 0 → 1 → 2 → 3 → 4 → 5. Follow every phase in order.

Prerequisite: The Atlassian Rovo MCP server must be available. Load references/confluence-mcp-tools.md for tool signatures and parameters.

Phase 0: Collect Parameters

Gather the required inputs before fetching any data.

Execution Mode Detection

Determine the execution mode before collecting any parameters:

1. Check CI env var: run echo $CI. If the value is true, set non_interactive = true.
2. Check skill arguments: if the skill was invoked with --non-interactive, set non_interactive = true.
3. Parse optional arguments:
   • --url PAGE_URL — fetch a specific Confluence page by URL
   • --space SPACE_KEY — browse pages within a specific space
   • --query SEARCH_TERMS — search Confluence for pages matching these terms

If neither condition is met, non_interactive = false (interactive mode — original behavior).

Collect Parameters

1. Get cloudId: Use mcp__confluence__getAccessibleAtlassianResources to retrieve the cloud ID. This is required for all subsequent Confluence API calls. If multiple sites are returned, ask the user to choose (interactive) or use the first one (non-interactive).

2. Gather search input:

   • Interactive: Check if the user provided a URL, space, or query as arguments. If not, ask:

     "How would you like to browse Confluence? You can:

     1. Provide a page URL to analyze a specific page
     2. Specify a space to browse pages within it
     3. Enter search keywords to find relevant pages

     What would you like to search for?"

     If the user provides a Confluence URL, extract the page ID from it. Common URL formats:

     • https://<site>.atlassian.net/wiki/spaces/<SPACE>/pages/<PAGE_ID>/<title>
     • https://<site>.atlassian.net/wiki/x/<shortlink>

   • Non-interactive: Use --url, --space, or --query value if provided. If none are provided, log a warning and exit gracefully:

     "Non-interactive mode: no --url, --space, or --query provided. Exiting. Provide at least one parameter to specify what to search."

3. Confirm before proceeding:

   • Interactive: Display and BLOCK until the user confirms:
     • URL mode: > "Fetching Confluence page at URL. Proceed?"
     • Space mode: > "Browsing pages in space SPACE_KEY. Proceed?"
     • Query mode: > "Searching Confluence for 'SEARCH_TERMS'. Proceed?"
   • Non-interactive: Log the parameters and proceed automatically:

     "Non-interactive mode: ."

Phase 1: Fetch Pages

Fetch pages based on the input mode determined in Phase 0.

URL Mode (specific page)

Use mcp__confluence__getConfluencePage with contentFormat: "markdown" to fetch the page content directly.

• If the page is not found or inaccessible, inform the user and stop.
• Result: a single page to analyze.

Space Mode (browse space)

1. If --space provides a space key, first use mcp__confluence__getConfluenceSpaces with the keys parameter to resolve the space ID.
2. Use mcp__confluence__getPagesInConfluenceSpace to list pages in the space, sorted by -modified-date.
3. Interactive: Display the page list and ask the user to select which pages to analyze (or "all").
4. Non-interactive: Take all pages (capped at 50).
5. For each selected page, use mcp__confluence__getConfluencePage with contentFormat: "markdown" to fetch full content.

Query Mode (search)

1. Use mcp__confluence__search with the user's query to find relevant pages.
2. From the results, identify Confluence page ARIs (filter out Jira issue results).
3. Use mcp__confluence__fetch or mcp__confluence__getConfluencePage to retrieve full content for each matched page.
4. Cap at 50 pages maximum. If more results exist, log a warning and process only the first 50.

Progress

• If no pages are found, inform the user and stop.
• Display progress: "Found N pages to analyze."

Phase 2: Extract & Filter Content

For each fetched page, extract and clean the content.

Content Extraction

• Use the markdown body content from each page.
• Record metadata for each page:
  • Page title
  • Page ID
  • Space (key and name)
  • Last modified date
  • Full Confluence URL (mandatory) — construct from the API response _links.webui field combined with the site base URL (e.g., https://<site>.atlassian.net/wiki<webui_path>). If the URL was provided as input, use it directly. Every page in the report MUST have a navigable URL — this is critical for source attribution in packmind-update-playbook.

Filtering

Apply these filters to discard noise:

Low-value filtering — discard pages that:

• Have empty or negligible content (less than 50 characters of text after stripping formatting)
• Are clearly non-technical based on title and content heuristics (e.g., meeting notes, HR policies, team events, vacation calendars, org charts)
• Are archived or marked as deprecated/outdated in their content

Title-based heuristics for non-technical content — discard pages whose titles strongly indicate non-coding content:

• Meeting minutes, standup notes, retrospective notes
• Onboarding checklists (unless coding-specific)
• Team directory, org charts, holiday calendars
• Budget, procurement, travel policies

Display progress: "Retained N substantive pages (after filtering)."

If no pages remain after filtering, inform the user and stop.

Phase 3: Classify for Relevance

Categorize each remaining page by playbook relevance.

High relevance (keep)

• convention — Documents a naming, structure, or organizational pattern (e.g., "API naming conventions", "folder structure guide")
• decision — Records an architectural or technical decision (e.g., "ADR: chose PostgreSQL over MongoDB")
• best-practice — Recommends a better approach for correctness, performance, or maintainability (e.g., "Error handling guidelines")
• recurring-pattern — Pattern or approach documented across multiple pages
• action-item — Commits to a specific change that affects coding practices or workflows

Low relevance (discard)

• non-technical — Content unrelated to software development
• process-only — Describes project management processes without coding implications
• outdated — Explicitly marked as deprecated or superseded by newer documentation

Retain only High-relevance pages.

Recurring Theme Detection

Group retained pages by semantic similarity. A recurring theme is 2+ pages that address the same underlying pattern. For each theme:

• Assign a short descriptive label
• List all contributing pages
• Note the occurrence count

Codebase Relevance Gate

After classification, apply a strict codebase relevance filter to all high-relevance findings.

Litmus test: "Would an AI coding agent need to know this when writing, reviewing, or shipping code in this repository?" — if no, discard the signal.

Signal	Verdict	Why
"All API endpoints must validate input with Zod"	KEEP	Coding convention
"Use feature flags for gradual rollouts"	KEEP	Architecture pattern
"Error handling: always wrap async calls in try-catch"	KEEP	Best practice
"Git branching strategy: trunk-based development"	KEEP	Dev workflow
"Team OKRs for Q1 2026"	DISCARD	Business metrics
"How to request PTO"	DISCARD	HR process
"Confluence space organization guide"	DISCARD	Tooling documentation
"Sprint planning process"	DISCARD	Project management

Edge cases: When a signal straddles both worlds (e.g., "How to set up your local dev environment"), apply the litmus test strictly — does it change how an AI agent writes or ships code? If it documents required environment variables, local setup commands, or build steps, KEEP. If it only describes how to install Slack or set up email, DISCARD.

Display progress: "Retained N codebase-relevant findings (discarded M non-relevant signals)."

Phase 4: Build Findings Report

Choose the output path based on execution mode:

• Interactive: .claude/tmp/confluence-review-findings.md
• Non-interactive: .claude/reports/confluence-review-findings-YYYY-MM-DD.md (using today's date, timestamped for CI artifact upload)

Create the target directory if needed.

Report Structure

# Confluence Documentation Review Findings Report

**Source**: <URL | Space KEY | Search "query">
**Pages analyzed**: N (after filtering)
**Actionable findings**: N
**Codebase-relevant findings**: N (of M high-relevance signals)

---

## Recurring Themes

### Theme: <label>
**Occurrences**: N pages

| Page | Space | Last Modified | Summary |
|------|-------|---------------|---------|
| [Page Title](url) | SPACE | YYYY-MM-DD | Summary of page content |
| [Page Title](url) | SPACE | YYYY-MM-DD | Summary of page content |

**Suggested playbook action**: <Create standard | Update standard X | Create skill | Create command | ...>
**Rationale**: <Why this theme warrants a playbook change>

---

## Individual Findings

### Finding: <short description>
- **Page**: [Page Title](url)
 : SPACE
 : YYYY-MM-DD
 : Condensed description of the coding practice or knowledge
 : Relevant excerpts from the page content
 : convention | decision | best-practice | recurring-pattern | action-item
 : 

---





| Signal summary | Reason discarded |
|---------------|-----------------|
|  |  |

URL Requirements

Every page reference in the report MUST include the full, navigable Confluence URL (e.g., https://mysite.atlassian.net/wiki/spaces/ENG/pages/123456789/Page+Title). This is critical because packmind-update-playbook extracts source URLs from the findings report for its packmind-cli diff --submit -m source attribution. The expected attribution format is:

<topic>: <summary> (source: Confluence <Page Title> https://mysite.atlassian.net/wiki/spaces/ENG/pages/123456789)

If a URL cannot be constructed (e.g., missing _links.webui), fall back to https://<site>.atlassian.net/wiki/pages/<pageId>.

Phase 5: Present and Hand Off

Interactive mode

1. Display a summary to the user:

   • Number of recurring themes found
   • Number of individual findings
   • List of suggested playbook actions

2. Ask the user:

   "Found N recurring themes and M individual findings. Review the full report at .claude/tmp/confluence-review-findings.md. Proceed to update the playbook with these findings?"

3. On confirm: Invoke the packmind-update-playbook skill with the full report content as the intent. This maps to Case B (explicit intent) of that skill's Phase 0. Frame the intent as:

   "Update the Packmind playbook based on the following Confluence documentation review findings report: "

4. On decline: Inform the user the report is available at .claude/tmp/confluence-review-findings.md for manual review.

Non-interactive mode

1. Display a summary (same as interactive: themes count, findings count, suggested actions).

2. Log the report output path:

   "Report written to .claude/reports/confluence-review-findings-YYYY-MM-DD.md."

3. Invoke packmind-update-playbook with the full report content as the intent, passing along non-interactive mode. Frame the intent as:

   "Update the Packmind playbook based on the following Confluence documentation review findings report: "

   The packmind-update-playbook skill will auto-approve all changes and delegate creation of new artifacts to the corresponding creation skills (packmind-create-standard, packmind-create-command, packmind-create-skill), all running in non-interactive mode.

Resources

references/

• confluence-mcp-tools.md — Atlassian Confluence MCP tool signatures and parameters used by this skill