Claude/repo analysis brainstorming e nf uw#2
Closed
alezenonos wants to merge 4 commits into
Closed
Conversation
Materializes the Solutions Exploration Confidence dimension as a written artifact (ideation.md) instead of a self-assessment. Forces ≥3 candidate approaches with trade-offs (or 1 + Single-Path Escape Hatch) and an explicit Selection before transitioning to PLAN. Rejected candidates feed RE-PLAN's ghost-constraint scan and can be reactivated when their rejection constraint turns out to be ghost. Validator enforces the gate when state ≥ PLAN. 12 new tests, 109 total. https://claude.ai/code/session_01WmjKGk8F6QgjpPDK49aBAn
Before generating candidates, the agent must scan classified constraints for any whose reclassification (hard ↔ soft ↔ ghost) would change which candidates are viable. If classification is uncertain on a viability- flipping constraint, the agent ASKs a focused question instead of guessing. If discovered mid-generation, the agent stops, re-classifies in findings.md, and re-runs the gate (state stays EXPLORE — the gate is an EXPLORE sub-step). This closes the gap where autonomous candidate generation produces a candidate set that doesn't include the right answer because a soft constraint was treated as hard (or vice versa). Process rule, no artifact change — validate-plan.mjs is unchanged. https://claude.ai/code/session_01WmjKGk8F6QgjpPDK49aBAn
Addresses two Codex review comments on PR #1: P1 — Empty/unreadable ideation.md silently passed validation. The early return on falsy readFile() result bypassed all ideation checks, letting a plan reach PLAN/EXECUTE without candidates or Selection. Now ERRORs with "ideation.md is empty or unreadable". P2 — Single-Path Escape Hatch was treated as populated when only "Why no alternatives" had content; "Falsification" being placeholder ("-") still passed. Both fields are protocol-required: a single-path claim without a falsification trigger has no STOP signal during EXECUTE. Now checks both, and emits a precise message when only one is populated. 3 new tests covering: empty ideation.md, escape hatch missing Falsification, escape hatch missing Why no alternatives. 112 tests, 0 failures (was 109). https://claude.ai/code/session_01WmjKGk8F6QgjpPDK49aBAn
Resolves conflicts that accumulated since PR #1 was opened: main has shipped the sub-agent architecture, Presentation Contracts (v2.17.0), SYSTEM.md atlas (v2.16.0), per-edit changelog (v2.15.0), plan-qualified DECISION anchors (v2.14.0), and other large changes. The original Ideation Gate was designed as an EXPLORE sub-step in a pre-sub-agent codebase; that no longer maps to the runtime. Re-homed for v2.18.0: - Gate now lives inside PLAN, owned by ip-plan-writer (not orchestrator). Writer is invoked in two phases: task=ideation produces ideation.md (>=3 candidates or 1 + populated escape hatch + Selection), task=plan writes plan.md after user approves the Selection. - New PC-IDEATION presentation contract (7th total). Emitted between PC-EXPLORE and PC-PLAN; user can approve, redirect, or request more options before plan.md is built. - Viability-flipping constraint check is now signaled by ip-plan-writer via NEEDS_USER_CLARIFICATION:<question> to the orchestrator, instead of an in-EXPLORE ask-the-user loop. - File Lifecycle Matrix: ideation.md row updated (W in PLAN by writer, R in EXECUTE, R+W in PIVOT for [REACTIVATED iter-N] markers). - PIVOT ghost-constraint scan: 4th question re-reads ideation.md Rejected list for candidates rejected against now-ghost constraints. Conflict resolutions: - VERSION: 2.10.0 (PR) and 2.17.0 (main) -> 2.18.0 - CHANGELOG.md: kept main's full history, rewrote PR's entry as v2.18.0 reflecting the architectural re-home (PC-IDEATION, two-phase writer, PIVOT-not-RE-PLAN terminology) - CLAUDE.md / README.md: kept main's rewrites; added ideation.md to plan tree, "Divergent ideation" row in Built-in Reasoning Frameworks table, badge bump to v2.18.0, test count 122 -> 134 - src/SKILL.md: kept main's structure; added Ideation Gate sub-step to PLAN section, ideation.md to File Lifecycle Matrix + filesystem tree + Mandatory Re-reads, split User Interaction PLAN row into "PLAN (ideation) -> PC-IDEATION" and "PLAN (plan) -> PC-PLAN", added 4th question to PIVOT ghost-constraint scan - src/scripts/bootstrap.test.mjs: kept both sets of tests; merged the bootstrap-files-exist list to include ideation.md AND changelog.md - src/references/file-formats.md, planning-rigor.md, bootstrap.mjs, validate-plan.mjs: auto-merged cleanly; PC-IDEATION definition added, ideation discipline rewritten to reference ip-plan-writer Runtime wiring added (not in PR or main): - src/agents/orchestrator.md PLAN dispatch: two-phase ip-plan-writer invocation with PC-IDEATION emission between phases, redirect/approve branching, contract count 6 -> 7 - src/agents/ip-plan-writer.md: dual-mode Your Task section, viability- flipping check at start of task=ideation, NEEDS_USER_CLARIFICATION signal, task-specific Output Format Tests: 135 passing (was 122 on main; +13 from the ideation suite). make validate passes. https://claude.ai/code/session_01WmjKGk8F6QgjpPDK49aBAn
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.