Learning Product

Learning Product

One Researcher Evidence Brief → one of four learning-product output types. The skill carries the brief's evidence into the format the audience can use, without weakening the rigour and without adding sources the brief did not verify.

When to use

Trigger on any of:

• "learning product", "learning brief", "MA brief from this Evidence Brief", "MA-facing brief"
• "workshop pack", "CoP pack", "facilitator pack", "build me a workshop from this brief"
• "regional learning brief", "federation brief", "regional synthesis"
• "donor memo from this brief", "management memo", "talking-points memo", "decision memo for donor"
• explicit naming of an output type (v1, v2, v3, v4, ma-brief, workshop-cop-pack, regional-learning-brief, donor-management-memo)

Do not trigger for:

• producing the underlying Evidence Brief — that is the researcher skill
• general decision memos not derived from a Researcher Evidence Brief — that is the decision-memo skill
• evaluation design or indicator design — those are separate skills
• ToC building — toc-builder

Required inputs

Ask in one batch. First two are required.

1. Evidence Brief slug (required): the directory name under literature-reviews/, e.g. 2026-05-10_cse-anti-gender-opposition-eca. The skill reads literature-reviews/<slug>/evidence-brief.md.
2. Output type (required): one of ma-brief (v1), workshop-cop-pack (v2), regional-learning-brief (v3), donor-management-memo (v4).
3. Recipient (required for v4 only): one of donor-programme-officer, donor-senior, ippf-en-management, ippf-central-management, mp-or-policy. Used to calibrate voice register and the decision-asked framing.
4. Audience focus (optional): MA cluster, sub-region, country list. Defaults derive from the Evidence Brief's geography metadata.
5. Length override (optional): word count or page count beyond default. Defaults defined per output type below.
6. Workshop format (optional, v2 only): single-90min (default), two-session-split, half-day, full-day. Affects slide count and segment timing.
7. Workshop modality (optional, v2 only): in-person (default), hybrid, virtual. Affects facilitator notes and microstructure adaptations.
8. Country snapshot count (optional, v3 only): default 4; allowed range 3–5.
9. Country naming convention (optional, v3 only): sub-region-label (default — Western Balkans MA, Caucasus MA, Baltic MA, Visegrád MA) or country-name (triggers redaction-candidate flagging).
10. Decision asked (optional, v4 only): the specific decision the recipient must make. If absent, the skill drafts a placeholder and flags it for replacement before publication.

Session start

1. Read this SKILL.md fully.
2. Read the Evidence Brief at literature-reviews/<slug>/evidence-brief.md.
3. Read the matching exemplar at agent-improvements/learning-product-exemplar-v[1|2|3|4].md for the requested output type. The exemplar is the structural target.
4. Read ~/.claude/CLAUDE.md audience-tier register section once per session if not already in context.

Workflow

Step 1 — parse inputs and route

Identify output type. If missing or ambiguous, ask Ane in one round:

"Which output type? ma-brief (2-page MA brief), workshop-cop-pack (workshop + CoP pack), regional-learning-brief (5–8 page federation brief), or donor-management-memo (1-page memo)?"

If donor-management-memo is selected without a recipient parameter, ask for it in the same round.

Step 2 — load source material

Read in order:

1. literature-reviews/<slug>/evidence-brief.md — the source content
2. literature-reviews/<slug>/sources.md if present — verified hyperlink list
3. literature-reviews/<slug>/literature-review.md if present — fuller synthesis for v3 depth
4. agent-improvements/learning-product-exemplar-v[N].md — structural target

The Evidence Brief frontmatter carries the region, period, and confidence metadata the output's title block uses.

Step 3 — apply the output-type sub-prompt

Route to one of the four sub-prompts below. Each sub-prompt instantiates the corresponding exemplar's structure against the Evidence Brief's content.

Step 4 — apply quality gates (all output types)

Before returning the draft:

• BLUF discipline. Sentence 1 = the verdict. Sentence 2 = the load-bearing reason. v1 headline ≤ 3 lines bold; v3 executive summary 5 bullets exact; v4 bottom line 1 paragraph bolding the decision-ask.
• Voice register. Tier 1 working brief by default. v4 unlocks directive register only at the decision-asked section. v2 facilitator notes use directive verbs for procedural steps.
• Citation placement. Tier 1 working brief: framework names off the running text; **Evidence base:** line at end of section. v2 source slide uses "For further details you can consult:" lead phrase per feedback_source_slide_framing.md.
• Citation rigour. Author surname + year + venue + section where applicable. Lifted verbatim from the Evidence Brief. Never invent.
• Hyperlink policy. Reuse only URLs that appear in the Evidence Brief or its sources.md. Never fabricate. Any URL not present in the source carries ⚠️ URL unverified — confirm before publication.
• Session-time URL audit step. Before returning the draft, verify each cited URL against the IPPF/UNFPA hyperlink standard:
• Aggregator wrappers (consensus.app, ResearchGate, academia.edu, Wikipedia) MUST be replaced with the canonical publisher URL or DOI before publication. Aggregators are tolerated only as supplementary alongside canonical, never as sole source.
• DOI links that resolve to a paywall MUST be supplemented with an open-access alternative (PMC, institutional repository, publisher-hosted open PDF, author manuscript) where one exists.
• Publisher-migrated URLs (Rowman to Bloomsbury, journal venue changes, etc.) MUST be updated to the current canonical URL.
• URLs returning 404 MUST be replaced with the canonical permanent link (typically UNESDOC, DOI, or institutional repository).
• When the canonical source has no open-access alternative and the audience is non-academic, the citation MUST flag "paywalled, no open-access alternative found" so MA staff and partner readers know in advance.

The audit applies WebFetch and WebSearch at session time. Lift the audit cost across the four output types (v1/v2/v3/v4) by running once per Evidence Brief, caching the URL substitutions, and reusing them across all output types generated from the same source brief.

• Sentence length. Under 25 words. Active voice. No em-dashes in body prose. No nominalisations where a verb works.
• Narrative prose model + fidelity gate (per agent-improvements/model-selection-policy.md). The illustrative narrative elements — v1 vignettes and v2 CoP workshop scenarios — MAY be drafted via a Fable per-call override (sanctioned: Fable was the more faithful drafter on narrative source-constrained prose, 2026-06-10). When any narrative element is Fable-drafted, the prose-fidelity gate (Gate 2 in the policy) is MANDATORY: run it on Sonnet against the source Evidence Brief and clear every flag before inclusion. The v4 donor / management talking-points memo and any persuasive or management-decision content stay on this skill's own model (Opus/Sonnet) — Fable is NOT used there, because the persuasive genre pulls both models toward overstatement; run the prose-fidelity gate with attention to overstatement and attribution inflation regardless.
• Plain English. Anglo-Saxon over Latinate. Acronyms spelled on first use. Translatability test for ECA / SSA / MENA readers.
• Disaggregation default. When the Evidence Brief carries WG-SS, age, gender identity, geography breakdown, the output preserves the disaggregation rather than collapsing.