# Principle Cultivation > Steward a project's fallible principle lineage: preserve source-bound research, turn only gate-passing inquiries into candidates, convene selective P-ID reviews, clarify interpretations, trial human-nominated alternates, and prepare human-only Sequence decisions. Use for reusable-principle research, candidate proposals, principle reviews, interpretation drift, alternate trials, or maintaining a principle-centered skill collection. - Skill: `lidessen/principle-cultivation` (Agent Skill, multi-file: 9 files) - Install (CLI): `npx skillmds@latest add lidessen/principle-cultivation` - Raw SKILL.md: https://api.skillmd.com/api/skills/lidessen/principle-cultivation/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: lidessen (https://skillmd.com/u/lidessen) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/lidessen/principle-cultivation --- # Principle Cultivation ## Principle expression **Primary:** P03 **Supporting:** P04, P12, P15 ## Principle source Use a host Sequence and interpretations when the host declares them. Otherwise use this package's read-only fallback in `references/sequence.md`. It contains the full one-line Sequence and the readings for P03, P04, P12, and P15, so this skill remains usable when installed alone. The fallback is a versioned projection, not a source to edit from a target project. ## What this stewards Treat `principles/SEQUENCE.md` as the only semantic source of core principles. The repository does not own truth: practice and source material remain outside it. It owns a compact, revisable lineage through which a later actor can recover why a principle was researched, proposed, reviewed, adopted, clarified, or left alone. ```text source-bound research ↓ when the candidate gate has plausible evidence candidate → selected review → human decision → Sequence → interpretation → expression ``` Research, candidates, adopted records, interpretations, and reviews are derivatives. None may redefine the Sequence or gain authority merely by being more detailed. A council or review member has an organizational role, not a second doctrine. Keep semantic admission, alternate participation, and standing-committee membership separate. A human-nominated candidate can gather bounded work evidence without becoming a P-ID, Primary, Supporting principle, current lead, or review member. ## Commands - `/principle-cultivation research ` — create or update one cited inquiry in `principles/research/`; `no-proposal` is a valid result. - `/principle-cultivation propose ` — apply the full candidate gate; create or update at most one linked candidate, never the Sequence. - `/principle-cultivation extract ` — deprecated compatibility path: research supplied evidence and create a candidate only when the same gate passes. Route open-ended inquiry to `research` instead. - `/principle-cultivation review ` — select a temporary P-ID team, pressure-test the target, and write review evidence. - `/principle-cultivation interpret ` — create or revise one source-bound interpretation without changing the Sequence. - `/principle-cultivation constitute` — run the exceptional founding review that nominates a standing committee; it never changes Sequence semantics itself. - `/principle-cultivation nominate-alternate ` — prepare the separate human decision for a bounded trial; it never implies adoption. - `/principle-cultivation trial ` — add one nominated candidate as a separately labeled trial seat and retain comparison evidence. - `/principle-cultivation adopt ` — prepare a human-approved sequence decision; append or revise a line only after explicit approval, then move its evidence to `principles/adopted/`. - No argument — inspect the lineage's live research, candidates, interpretations, and pending decisions; report the next useful action without writing. Read [references/source-citations.md](references/source-citations.md) whenever writing a source-bearing artifact. Read [references/research-template.md](references/research-template.md) for research, [references/candidate-template.md](references/candidate-template.md) for a proposal, [references/committee-protocol.md](references/committee-protocol.md) and [references/committee-review-template.md](references/committee-review-template.md) for review, and [references/interpretation-template.md](references/interpretation-template.md) for an existing P-ID's reading. ## Route by artifact and authority ### Research Use research when the user asks a durable question, supplies material worth preserving, or needs to compare theory with project evidence. Do not save every conversation: write `principles/research/.md` only on explicit durable request or an immediate proposal/interpretation handoff. 1. State the question and the decision it might change; distinguish neighboring concepts before naming a principle. 2. Keep descriptive inline links beside evidence-dependent claims. Prefer direct public sources or artifacts the user placed in scope; link rather than copy sensitive or licensed material, and record a limitation when an anchor is mutable or local-only. 3. Check existing P-IDs without extending them. Record what each already covers, the possible remainder, and the strongest `no-proposal` conclusion. 4. Give the note one stable slug and exactly one disposition: `open`, `no-proposal`, `candidate:`, `interpretation:P`, or `superseded:`. Update the same note on rerun; do not create duplicates. 5. A research note is revisable evidence. It cannot assign a P-ID, decide that a candidate is irreducible, or substitute for later candidate and review gates. ### Propose or extract Use `propose` only from a research record. The legacy `extract` path performs the same inquiry internally so existing callers retain its candidate-producing behavior when the gate passes. Before creating or updating a candidate, establish plausible evidence for every gate: - **Cross-context:** governs a class of choices, not one repository or anecdote. - **Decision-changing:** names a future choice that changes. - **Irreducible:** cannot be a straightforward combination of existing P-IDs. - **Falsifiable in use:** has a counterexample, boundary, or failed expression. - **Compact:** fits the one-line Sequence form without a hidden manual. - **Admission cost:** earns permanent attention rather than merely describing a useful local mechanism. If one gate is unresolved, keep the research `open` or `no-proposal`; never invent a candidate merely to preserve an attractive phrase. If it passes, create or update one `principles/candidates/.md`, link both artifacts, and leave the Sequence unchanged. ### Interpret Read the exact Sequence line first. A living interpretation may clarify shared reading, decision questions, source fidelity, boundaries, and expression probes. It must retain inline sources and stay character-for-character anchored to its P-ID line. If clarification adds an irreducible decision consequence or a new governing boundary, route to research and then proposal instead. ### Review and decide The default is to retain the current Sequence. For an addition, revision, retirement, or coherence check, follow `committee-protocol.md`: 1. Select 3–5 existing P-IDs by the present decision tension: lead, standing liaison, direct comparators, and a preservation seat. Do not summon all P-IDs. 2. Give each member the sequence, target, and evidence but not other reports. Require decision delta, overlap, boundary, unchanged alternative, and inline evidence. When delegation exists, assign independent agents; otherwise mark isolated passes self-reviewed. 3. Synthesize without counting votes. Preserve the Sequence unless recurrence, irreducibility, boundary, expression probes, and permanent attention cost all survive. Record dissent and the unchanged alternative. 4. Only the human may adopt, revise, retire, reject, nominate an alternate, or nominate standing members. Direct adoption remains valid but receives a retrospective review; the review never silently reverses it. ### Trial and reconcile For a human-nominated alternate, keep the normal Sequence P-ID team intact and add at most one separate trial seat. Before action, record baseline choice, candidate delta, and a disconfirming observation. Afterward append a sourced result—`strengthened`, `overlap`, `boundary failure`, or `failed`—and state whether existing P-IDs already explain it. No trial authorizes an irreversible, authority-changing, production, fact, or semantic action by itself. After human adoption, update the Sequence, move evidence from candidates to adopted, preserve stable IDs, and inspect only affected interpretations, skill lineages, portable snapshots, and L1 projections. An adopted record is history, not a second semantic source. ## Invariants - `principles/SEQUENCE.md` is citation-free, one stable principle per line, and the sole semantic source. - Research owns no P-ID or semantic authority. It has one stable slug and one declared disposition; candidates and reviews independently recheck its claims. - Candidate records never gain semantic authority; `principles/candidates/` holds only pending or incubating records, and `principles/adopted/` holds adopted evidence. - Every adopted P-ID has exactly one predictable interpretation path. It remains a revisable derivative and cannot redefine its source line. - Principle IDs are stable and never reused. A new or changed line requires explicit human approval and a dated selected review, retrospectively if needed. - A skill has exactly one Primary and at most three Supporting P-IDs. An alternate never enters lineage, leads work, sits on a sequence review, or enters a portable snapshot; one activation trials at most one alternate. - Source-bearing artifacts use descriptive inline links at the claims they support. A bibliography may supplement but never replace claim-level evidence.