Partnership Angles

Ground every one of the three library narrative templates (day-in-the-life, the-scale-gap, trust-travels) against the research and Connect-fit memo produced for a specific prospect, yielding three complete, citable, pitch-able angles. This skill is the terminal artifact of the propose-phase — it writes angles.yaml and stops; the orchestrator presents the angles to the operator and waits for a selection before proceeding to production. This skill runs inline at level 0; it does not dispatch Agent.

Inputs

Pipeline-only: a step of a /ace:partnership-video run; it reads that run's research.

Source	Artifact	Used for
partnership-research	ACE/partnerships/<slug>/research/deep-research.md	Cited org profile: scale, model, geography, expansion thesis
partnership-research	ACE/partnerships/<slug>/research/connect-fit.md	Connect/Dimagi capability-fit memo for this prospect
Phase 1 profile	ACE/partnerships/<slug>/prospect.yaml	Prospect identity, sector, target geography, contact
ACE repo (versioned)	templates/partnership-narratives/*/narrative.yaml	The three reusable narrative templates (durable library)

Products

• ACE/partnerships/<slug>/runs/<run-id>/angles.yaml — three grounded narrative angles (propose-phase terminal artifact)
• ACE/partnerships/<slug>/runs/<run-id>/run_state.yaml — Phase Write-Back: phases.angles.*

Process

1. Read the research artifacts and prospect context.

   Read the following from Drive via drive_read_file:

   • ACE/partnerships/<slug>/research/deep-research.md
   • ACE/partnerships/<slug>/research/connect-fit.md
   • ACE/partnerships/<slug>/prospect.yaml

   If either research file is missing, halt with an actionable error: "Partnership angles requires research/deep-research.md and research/connect-fit.md. Run partnership-research first."

   Extract from prospect.yaml:

   • name — display name of the prospect org
   • target_geography — the expansion thesis geography
   • sector — health / MNCH / agriculture / etc.
   • slug — used for all path construction below

2. Load the three narrative templates from the library.

   The library is at templates/partnership-narratives/ in the ACE repo. The three narrative IDs are:

   • day-in-the-life
   • the-scale-gap
   • trust-travels

   Each narrative.yaml defines: id, title, thesis, emotional_beat, hero, primary_capability, and seven beats (hook, cycle, handoff, scene, problem, product, impact), where each beat has an intent and a words budget.

   These templates are the source of truth for the arc structure. Do not invent narrative arcs. The skill's job is to ground the library intents with prospect-specific cited facts — not to author new narratives.

3. For each of the three narrative templates, produce a grounded angle object.

   For each narrative template, produce a grounded angle with the following fields:

   angle_id: <narrative-id>          # e.g. day-in-the-life
   title: <narrative title>
   logline: >                        # one sentence: what this angle says to THIS prospect
     <prospect-specific logline>
   thesis: <narrative thesis>
   hero: <narrative hero>
   emotional_beat: <narrative emotional_beat>
   primary_capability: <narrative primary_capability>
   beats:
     hook: <grounded one-liner: specific cited fact from the research that accomplishes the intent>
     cycle: <grounded one-liner>
     handoff: <grounded one-liner>
     scene: <grounded one-liner>
     problem: <grounded one-liner>
     product: <grounded one-liner>
     impact: <grounded one-liner>
   grounded: true | false
   grounding_notes: >               # cite the specific research fact(s) behind each beat;
     <per-beat source trace>        # if grounded: false, name the missing fact(s)
   

   Grounding rules (design §3, §8 — no inferred backstory):

   • A beat is grounded when a specific fact in deep-research.md or connect-fit.md directly supports what the beat's intent demands. The fact must be citeable (traceable to a source in the research report).
   • If a beat cannot be grounded — the research lacks the fact the intent requires — the beat text must say so explicitly: "[UNGROUNDED — missing: <description of the missing fact>]". Do NOT invent a plausible-sounding fact to fill it.
   • Set grounded: true only if all seven beats are grounded with real cited facts. Set grounded: false if any beat was left ungrounded, and name the missing fact(s) in grounding_notes.
   • A narrative that is grounded: false is still included in angles.yaml — the operator can decide whether to pursue additional research or skip it. Do not silently drop it.
   • Fabrication is the cardinal failure. A fabricated statistic, invented partnership, or invented program history in a prospect-facing artifact is the worst possible outcome. When in doubt: mark ungrounded, do not guess.

   Distinctness rule: The three grounded angles must lean on different research facts and different capabilities. If two angles ground the same beat with the same fact, the grounding is wrong — go deeper in the research to find differentiated evidence.

4. Write angles.yaml to the run folder.

   Resolve the run folder via drive_create_folder with findOrCreate: true (parent = the runs/<run-id>/ folder for this partnership run). Write angles.yaml via drive_create_file. Record the returned Drive file_id.

   The YAML structure:

   prospect_slug: <slug>
   generated_at: <ISO timestamp>
   angles:
     - angle_id: day-in-the-life
       ...
     - angle_id: the-scale-gap
       ...
     - angle_id: trust-travels
       ...
   

5. Inline QA (binary — run before writing the phase write-back).

   Verify all of the following. Record which checks pass and which fail — the write-back in step 6 is ALWAYS written regardless of outcome (the eval gate reads phases.angles.verdict from run_state.yaml; skipping the write-back would make the eval gate invisible to the orchestrator):

   • angles_file_exists: angles.yaml was written and a file_id was returned.
   • exactly_three_angles: angles.yaml contains exactly 3 angle entries.
   • angle_ids_match_library: The three angle_id values are exactly day-in-the-life, the-scale-gap, trust-travels.
   • all_seven_beats_present: Each angle has all 7 beat keys (hook, cycle, handoff, scene, problem, product, impact).
   • grounded_or_flagged: For every beat in every angle, either the text is a grounded one-liner OR the text contains [UNGROUNDED — no silent blanks.

6. Write the Phase Write-Back to run_state.yaml (always — both pass and fail paths).

   Write phases.angles.* 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 silently drops sibling keys; see CLAUDE.md § Gotchas):

   QA pass path (verdict: pass):

   phases:
     angles:
       status: done
       verdict: pass
       completed_at: <ISO timestamp>
       summary_artifact: angles.yaml
       steps:
         research_read: done
         narratives_loaded: done
         angles_grounded: done
         inline_qa: done
         write_back: done
       products:
         angles_file_id: <file_id from step 4>
         grounded_count: <0–3>    # number of angles where grounded: true
   

   QA fail path (verdict: fail): write the write-back first, then halt with the failed check names surfaced.

   phases:
     angles:
       status: incomplete
       verdict: fail
       completed_at: <ISO timestamp>
       summary_artifact: angles.yaml
       steps:
         research_read: done
         narratives_loaded: done
         angles_grounded: done
         inline_qa: fail
         write_back: done
       products:
         angles_file_id: <file_id from step 4>       # may be null if angles_file_exists check failed
         grounded_count: <0–3>
       qa_failures:
         - <failed-check-id>                          # one entry per failed check
   

   After writing the write-back on the fail path, halt with an actionable operator error naming the failed checks.

MCP Tools Used

• Google Drive: drive_read_file, drive_create_folder, drive_create_file, update_yaml_file

Note: The narrative library is read from the ACE repo filesystem at skill runtime — it is not stored in Drive. The three narrative.yaml files under templates/partnership-narratives/ are consumed as-is; no MCP call is needed to load them.

Mode Behavior

• Auto: Run all steps, ground all three angles, write angles.yaml and the phase write-back, stop. The orchestrator presents the angles to the operator; the operator selects one. No render happens in this skill.
• Review: Same as Auto. The propose-phase gate is at the orchestrator level (the operator picks an angle), not inside this skill.

Change Log

Date	Change	Author
2026-06-06	Initial version. Propose-phase terminal artifact skill for the partnership-video pipeline. Inline QA (no separate -qa skill), merge: deep write-back.	ACE team