Research

research workflow

Multi-source research workflow shipped with harnessed v2.0 (Stage ① Discuss); workflow.yaml schema bumped to harnessed.workflow.v3 in Phase v3.0-3.4 W1.1 (T3.4.W1.1 — D-09 L0 Discipline Substrate + D-05 tools_available cross-validate); T2.3 split 01-fan-out into 5 gated source lanes — each lane is now gated by one workflows/judgments/web-search-routing.yaml trigger (those 5 triggers used to be runtime orphans: zero gate ref, so the file was never loaded or evaluated). 02-synth unchanged.

Phases (6)

Lanes are mutually exclusive (subtask.search_type is single-valued); the default gateContext (search_type == 'keyword') fires only the Tavily lane, which is behaviourally identical to the pre-T2.3 unconditional single 01-fan-out.

Capability refs

Sister workflows/capabilities.yaml entries:

• tavily-mcp (impl: mcp-server, default 关键词)
• exa-mcp (impl: mcp-server, 描述式 / 学术)
• ctx7 (impl: cli-tool, 库 API 文档)
• gsd-discuss-phase (synth aggregate)

Invocation

• Slash command: /research <topic> (after harnessed setup)

Routing rules (bundled web-search routing — workflows/judgments/web-search-routing.yaml)

• 描述式查询 ("找一篇对比 X 和 Y 的博客") → Exa MCP
• 学术 / 论文 → Exa MCP
• 库 / API 文档 → ctx7 CLI
• 关键词 / 时效内容 → Tavily MCP (默认)
• 抓整站 / 站点结构 → Tavily crawl/map

How to invoke

!harnessed checkpoint intent research

The banner above (when present) means this invocation is REGISTERED with the engine (an intent marker) — not yet compliant: the steps below (prompt → spawn → checkpoint complete) resolve it, and a per-turn <workflow-intent> reminder persists until they run.

The numbered sequence below is the state machine — execute it with Bash. Do NOT improvise an equivalent flow from the Overview above: freelancing bypasses the engine (no ledger, no evidence guard). harnessed gives you the spawn-ready prompt; YOU spawn the subagent with a CC-native Task / Agent tool (keeps the session responsive + lets clarification round-trips reach the user).

Do NOT pipe to harnessed run research — that is the CI/headless path (in-process SDK spawn that blocks the session inside Claude Code).

1. Bash: harnessed prompt research --task "$ARGUMENTS" --json → parse {prompt, max_iterations, model}.
2. Spawn a CC-native subagent (Task / Agent tool) with that prompt and model, then drive delivery with harnessed's own completion gate:
   • on return, write the subagent's final output to a file and run harnessed checkpoint complete research --result-file <path> — it is fail-closed on the declared artifacts, the TDD boundary, and the verbatim <promise>COMPLETE</promise>.
   • if it blocks, run harnessed checkpoint fail research --failing-tests <n> to record the attempt; it prints BUDGET-EXHAUSTED / NO-PROGRESS / BREAK-LOOP when a stop condition is reached.
   • respawn ONLY while none of those three has fired. Any one of them means stop: re-scope the subtask, fix the blocker, or escalate to the user. Never respawn past a stop directive.
   • delivery contract: use a BLOCKING Agent/Task call — only a blocking call returns the subagent's final text as your tool result. A named/background teammate's final message is DISCARDED by the platform; if you must run it that way, instruct the agent to write its findings to a file (and read it back) or SendMessage them to the main session — otherwise the COMPLETE promise and the findings never reach you.
3. If the output contains STATUS: NEEDS_CLARIFICATION + a question list: STOP, relay them verbatim via AskUserQuestion, append the answers to the spec, then re-spawn the same sub.
4. On <promise>COMPLETE</promise>: write the subagent’s final output to a file, then Bash harnessed checkpoint complete research --result-file <path> --summary "<one-line>". Fail-CLOSED — it blocks unless every declared artifacts_expected file exists, the TDD boundary passes (non-empty evidence / both the red and green sides present / the test file was not deleted), and the result carries a verbatim <promise>COMPLETE</promise> (or a structured COMPLETE status). --result <text> is the inline variant; --result-file wins and is quoting-safe on Windows. --force records an audited override (evidence_status=overridden) — it does not silently pass.
5. If the complete gate blocked: Bash harnessed checkpoint fail research --failing-tests <n> to record the attempt. It prints BUDGET-EXHAUSTED / NO-PROGRESS / BREAK-LOOP once a stop condition is reached. Respawn ONLY while none of those three has fired; any one of them means STOP — re-scope the subtask, fix the blocker, or escalate to the user.