Skip to content

docs: capture itd-84 intent-decomposition discipline (draft) + SOTA note - #54

Merged
REPPL merged 1 commit into
mainfrom
docs/itd-84-intent-decomposition
Jul 13, 2026
Merged

docs: capture itd-84 intent-decomposition discipline (draft) + SOTA note#54
REPPL merged 1 commit into
mainfrom
docs/itd-84-intent-decomposition

Conversation

@REPPL

@REPPL REPPL commented Jul 13, 2026

Copy link
Copy Markdown
Owner

What

Captures itd-84 — a draft discipline-kind intent for the capture-time
intent-decomposition validator discussed in the auto-merge review — plus the
SOTA research note it is drawn from.

The rule: at /abcd:intent capture, before an intent is filed, a validator
produces an advisory analysis a human confirms —

  1. decompose each part to its record home (capability → intent, trust-rule →
    ADR + invariant, stance → principle, plumbing → brief);
  2. surface interdependencies via a deterministic candidate pass, with typed
    cross-links (supersedes/reverses/duplicates/refines);
  3. a reversal is only ever flagged, never auto-classified or auto-filed.

Rooted in SOTA (per sota-per-intent)

The design is not hand-waved — a sota-researcher run
(research/notes/2026-07-13-intent-decomposition-sota.md) gives an adopt/adapt/
bespoke call per component:

Half Maturity Call
Deterministic candidate-finding Mature adapt-with-seam (Go lexical; embeddings optional)
Semantic decomposition Usable adapt-with-seam (host-delegate to the agent fleet)
Reversal/contradiction reasoning Research-only bespoke-with-seam, advisory-only, never auto-file
Advisory-gate mechanics Practitioner-proven adopt (3 outcomes, sample-size floor, safety veto, per-version windowing)

No bespoke no-seam build anywhere → clears abcd's hard stop.

How it fits the record

  • MVP-first per the promotion ladder: the principle + hand-run protocol + a
    deterministic Go pre-pass are the first rung; the automated agent follows.
  • The agent inherits itd-81 (calibrated, never blocking).
  • Subsumes the unbuilt intent_sota gate itd-29 gestures at — enforcing
    SOTA-rooting is one check it runs.
  • Embodies facilitator-default-thinker-optional (in PR docs: auto-merge record groundwork (decision + facilitator/thinker principle) #53, referenced in
    prose here to avoid a cross-branch link).

Status / checks

Draft, awaiting maintainer adoption — placed in intents/disciplines/ with
staged (loud) enforcement; move to a different rung if you'd rather it not read as
active. record-lint: 0 blockers, no new findings. docs-lint: 0 findings. No code.

Branched from main, independent of PRs #50#53.

Assisted-by: Claude:claude-opus-4-8

…note

A discipline-kind intent for the capture-time intent-decomposition validator:
before an intent is filed, a proposed capability is decomposed into its record
homes (experience -> intent, trust-rule -> ADR + invariant, stance -> principle,
plumbing -> brief) and its interdependencies surfaced, as an advisory analysis a
human confirms. Rooted in SOTA (2026-07-13 research note): deterministic
candidate-finding (adapt-with-seam), semantic decomposition (host-delegate to the
agent fleet), reversal detection strictly advisory (research-only, over-flags),
adopt the three-outcome advisory-gate pattern. Build is MVP-first per the
promotion ladder; the automated agent inherits itd-81 and never blocks. Subsumes
the unbuilt intent_sota gate; embodies facilitator-default-thinker-optional.

Draft awaiting maintainer adoption. record-lint + docs-lint clean.

Assisted-by: Claude:claude-opus-4-8
@REPPL
REPPL merged commit 4dff770 into main Jul 13, 2026
12 checks passed
@REPPL
REPPL deleted the docs/itd-84-intent-decomposition branch July 13, 2026 08:22
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant