mk:confluence-search

Forks to the confluence-search agent. CQL is positional on search (and on validate / build helpers); flags everywhere else. The agent enforces CQL sanitization for any user-derived term via scripts/cql-sanitize.sh.

Triggers

• "search confluence for ..."
• "find pages where space = ENG and title ~ 'roadmap'"
• "build a cql for ..."
• "export the search to /tmp/out.csv"
• "list / get / favourite a saved filter"
• "list spaces" / "get space ENG"

Examples

• "find pages updated in the last 7 days in space ENG, top 20"
• "validate this cql: space = ENG AND title ~ 'Q3'"
• "list every space I can see"
• "save a filter named 'My recent specs' with this cql: ..."

See also

• Agent: ../../agents/confluence-search.md
• Shared: ../confluence/references/{install-and-auth,cli-idioms,safety-framework}.md
• Domain refs:
  • references/cql-patterns.md — canonical CQL templates (curated)
  • references/cql-reference.md — full CQL operator + function reference
  • references/search-examples.md — practical query examples per scenario
• Peer leaves: mk:confluence-bulk (write-side bulk by CQL — same dry-run discipline), mk:confluence-spec-analyst (read-only deep analysis), mk:confluence-page (single-page CRUD)

Gotchas

• confluence-as does NOT export _escape_cql_string — it lives as a private function in the search command module. The cql-sanitize.sh shared script is the only safety gate. Use it unconditionally for any user-derived CQL term. [from research]
• The --quiet global flag is declared in confluence-as but unimplemented (never read in main). Don't rely on it; redirect stderr explicitly when needed. [from research]
• Keychain entry name "confluence-assistant" collides with upstream Confluence-Assistant-Skills if both tools are installed. We never use keychain (env-only via wrapper) so we're not affected — but document the collision so users running both know. [from research]
• Two-tier filesystem cache (SkillCache SQLite + ~/.confluence-skills/cache/) is opaque. Don't depend on cache state — confluence-as ops cache-clear is safe to invoke if behavior looks stale. [from research]
• Grow this list as new edge cases surface.