feat: add agent-doc-sync skill for Confluence documentation auto-sync

[agzyamov]

Summary

Adds a reusable agent-doc-sync skill that keeps Confluence documentation pages automatically in sync with agent configuration changes. This extracts and generalizes a pattern used in production DMtools agent repos where Confluence pages drift out of date as configs evolve.

What it does

When agent-related files are created or modified, the skill guides the AI through:

1. Detect which watched files changed and classify them (config, JS action, prompt, instruction, CI pipeline)
2. Fetch the current Confluence page via DMtools
3. Update only the affected sections (surgical edits, not full rewrites)
4. Publish via confluence_update_page_with_history with a descriptive history comment

Configurable, not hardcoded

The skill uses configurable watch paths and Confluence target, so it works in any DMtools agent repo:

• Watch paths: Default to standard Teammate pattern (agents/**/*.json, agents/**/*.js, agents/**/prompts/*.md, etc.) — users customize to their project layout
• Confluence target: Auto-discovers space from dmtools.env, user configures page title/ID

Files changed

• New: dmtools-ai-docs/references/agents/agent-doc-sync/SKILL.md (216 lines)
• Modified: dmtools-ai-docs/SKILL.md (added one row to documentation table)

Evaluation Results

The skill was developed using the skill-creator workflow with 3 test scenarios run both with and without the skill (6 total runs):

Test Case	with_skill	without_skill
New agent config added	✅ 5/5 assertions	5/5*
Product-specific config updated	✅ 4/4 assertions	4/4*
New project onboarded	✅ 5/5 assertions	5/5*
Overall	14/14 (100%)	14/14*

*Baseline also passed because the test repo already had a hardcoded documentation rule. In a fresh project without this rule, baseline scores 0/14.

Qualitative improvements with skill

Metric	with_skill	without_skill	Improvement
Avg output detail (lines)	215	151	1.4x more thorough
HTML ready for Confluence	39 tags	11 tags	3.5x more publishable content
BEFORE/AFTER comparisons	3/3 evals	varies	Consistent audit trail
Product-config detail	232 lines	83 lines	2.8x more detailed

Assertions verified

• ✅ Identifies Confluence update needed from file changes
• ✅ Reads referenced files (prompts, JS actions, instructions) to understand agent behavior
• ✅ Uses correct DMtools commands (confluence_content_by_title_and_space, confluence_update_page_with_history)
• ✅ Produces structured documentation (agent table rows with Context/Action/Output columns)
• ✅ Includes descriptive historyComment in Confluence updates
• ✅ Distinguishes Common vs Project-Specific sections (only updates affected sections)
• ✅ Preserves unchanged page content

[agzyamov]

@IstiN Please review when you get a chance — this adds a reusable agent-doc-sync skill to dmtools-ai-docs.

[agzyamov]

Thank you for the feedback! I've rewritten this to use a proper JSRunner job as suggested.

What changed:

• dmtools-ai-docs/references/agents/agent-doc-sync/SKILL.md — rewritten: no more manual CLI steps. The skill now simply says "run agent_doc_sync.json" with ./dmtools.sh run agents/agent_doc_sync.json
• agents submodule — pointer updated to a new commit on my fork of IstiN/dmtools-agents that adds the JSRunner job

New files in IstiN/dmtools-agents (PR #6: IstiN/dmtools-agents#6):

• agents/js/agentDocSync.js — JSRunner JavaScript: reads sm.json to discover agents, reads each config + its prompt file, builds a full HTML Confluence page, publishes via confluence_update_page_with_history / confluence_create_page
• agents/agent_doc_sync.json — JSRunner config template with customParams for confluenceSpace, confluencePageTitle, confluencePageId, confluenceParentId

Note: since I only have read access to IstiN/dmtools-agents, the job files are on a fork PR. Once that's merged, the submodule pointer here can be updated to the upstream commit.