agentleFS
egregore-labsegregore-labs/egregore/.claude/skills/harvest/SKILL.md

harvest

Skill293 starsChanged 11 days agoMITView on GitHub

Run an adaptive harvest — directed elicitation that extracts, deepens, and synthesizes what people think about a topic. Use for 'harvest', 'run a harvest', 'align the team on', or a structured interview; not for a casual question or fixed survey.

17 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 14 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run an adaptive harvest — directed elicitation that extracts, deepens, and synthesizes what people think about a topic.

    line 13
  2. Distinctive:

    Read these completely before questioning:

    line 17
  3. Very specific:

    Use AUDIT.md §14 only when a pasted decision-surface return requires its review grammar.

    line 23
  4. Very specific:

    If a named gap remains, make one bin/search.sh query episode and open only the missing canonical evidence.

    line 32
  5. Distinctive:

    Ask one interaction at a time and always allow freeform correction.

    line 44

“Common” is measured: kept by at least 6 in 10 of the 14 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Harvest

Elicit tacit preferences, positions, constraints, and knowledge, then preserve an attributed synthesis. A harvest maps choices; it does not force convergence.

When to invoke

Run an adaptive harvest — directed elicitation that extracts, deepens, and synthesizes what people think about a topic. Use for 'harvest', 'run a harvest', 'align the team on', or a structured interview; not for a casual question or fixed survey.

Load the cognitive contracts

Read these completely before questioning:

  • .claude/skills/harvest/PROCESS.md — adaptive elicitation rhythm
  • .claude/skills/harvest/QUESTION_PALETTE.md — intent, move, answer shape
  • .claude/skills/harvest/FORMAT.md — evidence-bound synthesis grammar

Use AUDIT.md §14 only when a pasted decision-surface return requires its review grammar. Its historical graph/file persistence recipes are not an execution contract; Runtime adapters below own persistence.

Seed

Parse topic plus optional --respondents, --seed, --resume, and --mode blind|disclosed|comparative. Reuse attached EGREGORE_ORG_CONTEXT_V1; do not repeat retrieval or reopen supplied sources. If a named gap remains, make one bin/search.sh query episode and open only the missing canonical evidence. Do not query infrastructure directly.

Make low-confidence RoleSheet assumptions visible. In blind mode, other respondents' answer content neither appears nor shapes their questions. For a blind shared-artifact round, dispatch the frozen question set unchanged.

Elicit

Apply PROCESS.md's seed → generate → evaluate → checkpoint → cascade → synthesize rhythm:

  • Ask one interaction at a time and always allow freeform correction.
  • Record a questionIntent and evidence-bound evaluation for each answer.
  • Watch for satisficing and sycophancy; name real tensions.
  • Offer checkpoints at natural transitions and respect “I'm done.”
  • Attribute disclosed positions. Never turn a majority into anonymous team alignment.
  • While collection is open, a respondent may correct their answer. After the declared completion condition seals the evidence, a changed position starts a new round; there is no automatic second pass.

For an absent respondent, create one authorized canonical question per turn:

bash bin/question.sh create \
  --from "{initiator}" --to "{respondent}" \
  --topic "{topic}" --question "{question}" \
  --harvest-id "{harvest_id}" \
  --harvest-session-id "{harvest_session_id}" \
  --turn "{turn}" --question-intent "{intent}" \
  --context-mode "{blind|disclosed|comparative}"

This records the pending question; it does not notify anyone. Delivery requires a separate action permission and exact recipient/message consent.

Decision-shaped rounds

For three or more genuine interrelated forks, use the existing Meridian decision-surface renderer and QUESTION_PALETTE modes. Self surfaces may remain local. Do not publish, share, or deliver a directed surface implicitly. A separate permitted, explicitly consented action is required. Absorb returned answers into the evidence set; keep UNDECIDED open. Do not write graph or event-log projections directly.

Synthesize and write once

Build the attributed layered synthesis from FORMAT.md. Every assertion must trace to a recorded answer or marked seed. Show the synthesis and wait for acceptance or edits. Then render one complete Markdown body to a temporary file under tmp/ and make exactly one canonical synthesis write; {synthesis-file} is that file, {synthesis-title} is the accepted topic followed by — harvest synthesis, and {topic} and {workstream} come from the accepted synthesis:

bash bin/knowledge.sh create \
  --type finding --subtype harvest-synthesis \
  --title '{synthesis-title}' \
  --input '{synthesis-file}' \
  --topic harvest --topic '{topic}' \
  --workstream '{workstream}'

The Runtime owns ActorContext authorization, stable provenance, Markdown/Git, retrieval refresh, background embedding, and content-free telemetry. Do not run a second save, index, graph, publish, share, or notification action.

Render the compact ⊙ HARVEST completion with topic, respondents, disclosure mode, canonical path, and writeback status. Canonical Markdown wins over every optional projection.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.