partnership-research

name: partnership-research description: > Research a non-Connect prospect org for a partnership video: deep web research (what they do, scale, model, geography, the expansion thesis) plus a Connect/Dimagi capability-fit memo. Cited. disable-model-invocation: true

Partnership Research

Research a non-Connect prospect organization in preparation for the partnership video pipeline. Produces two verified, cited artifacts: a deep web research report (org profile, scale, model, geography, expansion thesis) and a Connect/Dimagi capability-fit memo grounding what Connect specifically unlocks for this org type in the target geography. This skill runs in the research phase and is dispatched by the level-0 partnership-video procedure doc.

Inputs

Products

• ACE/partnerships/<slug>/research/deep-research.md — cited org profile from deep web research
• ACE/partnerships/<slug>/research/connect-fit.md — Connect/Dimagi capability-fit memo
• ACE/partnerships/<slug>/runs/<run-id>/run_state.yaml — Phase Write-Back: phases.research.*

Process

1. Read prospect.yaml to establish the research scope.

   Read ACE/partnerships/<slug>/prospect.yaml via drive_read_file. Extract:

   • name — prospect organization's display name
   • current_program — what they do today
   • target_geography — the expansion thesis geography
   • sector — health / MNCH / agriculture / etc.
   • contact — who we are in discussions with

   If prospect.yaml is missing or unreadable, halt with an actionable error: "Partnership research requires ACE/partnerships/<slug>/prospect.yaml. Run the profile phase first or create the file manually with required fields."

   If a --prospect-folder Drive folder ID was supplied by the operator, also read its contents via drive_list_folder then drive_read_file on each listed file. Treat these as highest-signal context. Do NOT halt if the folder is empty or inaccessible — log a warning and proceed from prospect.yaml alone.

2. Ensure the research folder exists.

   Resolve ACE/partnerships/<slug>/research/ via resolve_opp_path to get the partnership root, then drive_create_folder with {name: 'research', parentFolderId: <partnership-root-id>, findOrCreate: true}. Retain the returned folder ID for writes in steps 3 and 4.

3. Deep web research: invoke the deep-research skill.

   Using the prospect identity and target geography, compose a refined research query:

   "What does [org name] do, at what scale, in which geographies, using what delivery model? What is their expansion thesis for [target geography]? What is their evidence of impact and their published funding/programmatic commitments?"

   Weave in any high-signal context from the operator's --prospect-folder to sharpen the query. Invoke the deep-research skill — the orchestrator runs at level 0, so deep-research's internal Agent dispatches are legal.

   The output must include:

   • Org overview: what they do, for whom, geographic footprint
   • Scale: delivery numbers, FLW/worker count, beneficiary reach
   • Delivery model: how work is organized and verified
   • Expansion thesis: documented or inferred interest in the target geography
   • Evidence of impact: published citations, evaluations, donor reports
   • Citations section with ≥3 sourced URLs

   Write the cited report to research/deep-research.md via drive_create_doc_from_markdown (parent = research folder from step 2). Record the returned Drive file_id.

   Guardrail: Every factual claim must trace to a cited source. Unverifiable claims must be flagged explicitly: "Unverified — could not locate a source for this claim." Do NOT invent statistics, program histories, or partnership relationships. If a fact cannot be sourced, omit it or flag it.

4. Build the Connect/Dimagi capability-fit memo.

   Cross-reference Connect/Dimagi capabilities against the research output and real ACE PDDs, program library artifacts, and Dimagi case studies that are accessible in Drive under ACE/. The goal is a grounded answer to: "What does Connect specifically unlock for this org type in the target geography that they cannot easily do today?"

   Capability validation rules (design §8 "close the loop"):

   • Every stated Connect capability must be validated against real artifacts (PDDs, atom schemas, program outputs in Drive), not asserted from memory.
   • Browse or read relevant PDD examples via drive_list_folder on ACE/ to find analogous programs for the org's sector and geography.
   • Cross-reference against documented Connect features (Learn→Deliver→Verify→Pay loop, payment-for-verified-delivery, rapid program stand-up, funder-grade reporting, mobile-first FLW tools) anchored to specific real programs.

   Structure the memo:

   • What they do today (gap): What the org currently relies on, and what limitations exist.
   • What Connect unlocks: Per-capability bullets, each citing a real analogous ACE program or documented Connect feature as evidence. Minimum 3 concrete capabilities.
   • Expansion fit: Specifically why Connect is well-suited for the target geography (infrastructure, payment, verification needs).
   • Risk factors: What would need to be true for Connect to be the right fit (scale, LLO availability, connectivity).

   Write to research/connect-fit.md via drive_create_doc_from_markdown (parent = research folder from step 2). Record the returned Drive file_id.

5. Write the Phase Write-Back to run_state.yaml.

   Write phases.research.* to ACE/partnerships/<slug>/runs/<run-id>/run_state.yaml via update_yaml_file with merge: 'deep' (a partial nested patch of phases.<phase> requires deep — two-level would silently drop sibling keys; see CLAUDE.md § Gotchas):

   phases:
     research:
       status: complete
       verdict: pass
       completed_at: <ISO timestamp>
       summary_artifact: research/deep-research.md
       steps:
         prospect_read: done
         deep_research: done
         connect_fit: done
         write_back: done
       products:
         deep_research_file_id: <file_id from step 3>
         connect_fit_file_id: <file_id from step 4>
   

   If either research file was not written (e.g., deep-research errored), set verdict: fail and status: incomplete, surface the failure reason, and halt before writing complete.

MCP Tools Used

• Google Drive: drive_read_file, drive_list_folder, drive_create_folder, drive_create_doc_from_markdown, resolve_opp_path, update_yaml_file
• Skills (via Skill tool, dispatched from level-0 orchestrator): deep-research

Note: deep-research is a skill dispatched via the Skill tool — it is not an MCP atom. The level-0 orchestrator is what calls it, making the Agent fan-out legal per ACE topology rules.

Mode Behavior

• Auto: Run all steps, write both research artifacts and the phase write-back, proceed. Flag any unverifiable claims inline in the artifact.
• Review: Same as Auto. No human pause point — the downstream partnership-research-eval surfaces quality concerns for human review after the artifacts are written.

Change Log