cardboard-box

Phase A (Abstract) — AgentFold compression of the Locate trajectory into a ≤2000-token working-memory summary plus a Memex excerpt store. Runs in a clean-context subagent to prevent trajectory contamination. Mechanically validates that every FINDING-XXX, GAP-XXX, and ESCALATION_TRIGGER is preserved. Use after findings.md is complete and before synthesis. Triggered automatically at phase boundary, or at ≥60% context-used telemetry threshold.

Phase L (Locate) — bounded probes and scatter subagents to answer DECISION_TARGET sub-questions. Descends a symbol → graph → lexical → windowed-read → test-dry-run ladder; emits FINDING-XXX records with evidence anchors and confidence tiers. Hot phase (≤60% of mission budget). Use after map.md is complete and sub-questions remain. Trigger phrases — "find where X happens", "trace Y", "who writes to Z", "who calls this", "what implements this".

Incremental delta re-scout (Phase A/T mode) — a READ-ONLY, evidence-anchored re-run that reuses a prior scout-report + its Memex store + a git-diff range to re-probe ONLY the changed surface, carrying unchanged findings forward verbatim. Narrows (does NOT close) the always-on-live-index staleness gap. Use at Phase A intake when a prior scout-report.md + Memex store exist for the same surface and the question is "what changed since the last scout" or "re-verify the prior map against HEAD". Do NOT use for a first-time scout (run the full A→T→L→A→S cycle instead).

Phase L sub-mode (Scatter-Gather Locate) — TRANCE-gated parallel fan-out of read-only Locate probes across topologically-disjoint module clusters, each in a clean-context subagent, merged back into findings.md via a deterministic dedup contract. Never default; serial Locate is the standard path. Use ONLY when map.md surface is large (>5 modules OR >25 files) AND ≥2 disjoint DECISION_TARGET sub-questions exist. Trigger phrases — "audit across all modules", "trace these N independent flows", "scout this large surface in parallel". Do NOT load for a single sub-question, tightly-coupled sub-questions, or a tight budget.

Phase S (Synthesize) — emit the scout-report.md artifact that downstream agents (SPECTRA, APIVR-Δ, human) consume. Hard cap 3000 tokens; every factual clause cited with FINDING-XXX; every recommended action carries a handoff label. Use after Phase A (Abstract) fold is validated and DECISION_TARGET has an answer or its gaps are recorded. Final phase of an ATLAS mission.

Phase T (Traverse) — deterministic structural mapping of a codebase. Zero LLM calls during retrieval; only symbol index, AST, ripgrep, and git log. Emits map.md with entrypoints, modules, call graph, and heatmap. Use after mission.md is valid and before any meaning-based search. Trigger phrases — "map the repo", "what are the entrypoints", "build the call graph", "list all routes/workers/CLIs".

Receiver-side ECL integrity gate (blocking, symmetric). Refuses to process any upstream artefact whose sibling .envelope.json lacks a verify_pass trace event from the orchestrator. Use when reading any upstream artefact handed off to ATLAS that carries a sibling .envelope.json. Symmetric across all Eidolons — every Eidolon enforces it identically per ECL §6.2.2.

Governs hypothesis generation, adversarial stress-testing, and multi-criteria scoring during the FORGE Reason phase. Use when the Reasoner enters Phase R to generate ≥3 distinct hypotheses, run Inversion/Boundary/Pre-Mortem/Dependency stress tests, score on a 5-dimension weighted rubric, and execute multi-pass deliberation. Do not use when Self-Consistency Mode (G2/TRANCE) is active — load forge-self-consistency instead.

Governs problem decomposition and constraint extraction during the FORGE Frame phase. Use when the Reasoner enters Phase F to extract the core decision, classify its type, map hard/soft constraints, set deliberation depth, and declare success criteria before any hypothesis work begins.

Governs N perspective-diverse, mutually-blind reasoning traces with structural-agreement judge-merge for gated high-stakes FORGE decisions (G2/TRANCE). Use instead of forge-deliberation when Deep depth AND a stakes flag both hold (Framing score 8–9 with irreversible/system-wide blast radius), or on explicit opt-in. Do NOT use for standard-tier decisions — self-consistency is never the default.

Governs logic verification, evidence coverage, confidence calibration, and the REFORGE protocol during the FORGE Gate phase. Use when the Reasoner enters Phase G to check logical soundness (fallacy scan), verify evidence coverage, assess decision completeness, calibrate the 4-factor confidence score, and decide pass/emit or trigger a single REFORGE revision pass.

Blocking receiver integrity gate for inbound ECL hand-offs to FORGE (ECL §6.2.2). Use automatically when reading any upstream artefact that carries a sibling `.envelope.json` — verifies `verify_pass` trace event keyed by `message_id` before processing the payload. Refuses and returns control to the orchestrator on integrity mismatch, missing event, schema violation, or undeclared edge. Do not use for artefacts without a sibling envelope.

Governs how the Scribe transforms context into structured, grounded prose during the Draft phase. Use when starting any document composition, loading templates, applying structural markers, or writing section by section with topological ordering and envelope-aware intake.

Operationalizes the G5 gated parallel doc-section synthesis mode for IDG. Use when composing a large document (six or more independent sections) routed at the TRANCE tier; fires dependency-layering, bounded per-layer fan-out (up to five subagents), per-section CHT mini-gate, topological-order parent assembly, and provenance merge. Do NOT use for standard-tier or small documents — sequential composition is always the default.

Defines the CHT verification framework and provenance output for the Gate phase of the IDG cycle. Use when entering the Gate phase, verifying document completeness and truthfulness, running CHT scores, deciding DELIVER vs REVISE vs ESCALATE, or producing provenance metadata blocks.

Blocking symmetric receiver integrity gate for inbound ECL hand-offs to IDG. Use automatically when reading any upstream artefact with a sibling .envelope.json — verifies SHA-256 integrity via orchestrator trace, validates contract conformance, and refuses (does not process) on any failure. Do NOT use for non-ECL artefacts or when no envelope sidecar is present.

Phase K triage gate that decides whether Kupo will attempt a delegated micro-task or refuse cheaply. Use when a delegation arrives and Kupo must decide KEEP vs REFUSE/ESCALATE; runs once per delegation at the start of every KUPO cycle. Do NOT use during Phase U/P/O.

Phase P+O sandbox patch loop with external verifier; governs edit emission, harness applier invocation, per-file loop detector, circuit-breaker, and pre-completion green-signal gate. Use when Kupo enters Phase P to emit a patch or Phase O to run verifiers; load after a KEEP decision. Do NOT use during Phase K or U.

Blocking ECL envelope integrity gate for inbound Kupo hand-offs; verifies SHA-256 and contract conformance before any payload is processed. Use when an upstream artefact arrives with a sibling .envelope.json; load automatically on every inbound ECL hand-off. Do NOT use for non-ECL artefacts (no .envelope.json sibling).

Open-ended requirements & stakeholder/goal elicitation BEFORE CLARIFY. Use when the GOAL itself is underspecified (IDEA/STRATEGIC intent) — distinct from CLARIFY, which disambiguates an already-known goal. Read-only, bounded; produces an elicitation summary, never a plan.

TRANCE-only G3 evaluator-optimizer mode: fan out 2-4 perspective-diverse clean-context candidate specs, score with a bias-hardened evaluator, judge-merge into ONE spec, terminate at confidence ≥85% or hard cap 3 iterations. Read-only; never the default — runs ONLY when the cortex authorizes TRANCE.

Decision-ready specifications via the SPECTRA cycle. Use for complex features (complexity ≥7/12), multi-component/service changes, ambiguous requirements needing structured decomposition, or any task where 'just start coding' would require significant rework. Produces a dual-format spec (Markdown + YAML/JSON), never code.

Load when reading any upstream artefact handed off to SPECTRA that carries a sibling .envelope.json. BLOCKING per ECL §6.2.2: the orchestrator MUST have verified the envelope's SHA-256 (eidolons verify-envelope --block / eidolons run --verify) and recorded a verify_pass before dispatch. If no verify_pass exists for the message_id, or a verify_fail is present, REFUSE to process the payload and hand back to the orchestrator. Symmetric receiver gate — every Eidolon enforces it identically.

Phase G (Graph) — builds the Information Dependency Graph from candidate fault nodes. Distinguishes propagated symptoms from root candidates using information flow, not temporal order; ranks root candidates by descendant count. Use after `fault-surface.md` is schema-valid with ≥1 candidate, before the Intervene phase.

Phase I (Intervene) — falsifies hypotheses via counterfactual replay. Designs minimal interventions for each root candidate; runs in sandbox; root cause is the candidate whose intervention flips failure→success. Hard cap 5 interventions. Use after `idg.md` identifies ≥1 ROOT_CANDIDATE, before emitting the root-cause report.

Phase I (Isolate) — narrows the fault surface to ≤8 candidate nodes via delta-debugging-style reduction. Preserves failure throughout reduction; records ruled-out evidence. Use after `reproduction.md` is stable and `DETERMINISM_VERDICT ≠ intermittent`, before the Graph phase.

Phase L (Learn) — emits verified root-cause finding with evidence chain, walks the cause back to its originating decision, persists failure signature to semantic memory for future de-duplication, routes handoff to the correct downstream Eidolon. Use after `intervention-log.md` has a SURVIVOR to produce the primary deliverable.

Blocking, symmetric receiver-side integrity gate for inbound ECL hand-offs. Enforces ECL §6.2.2: REFUSES to process any upstream artefact whose SHA-256 envelope has not been verified and passed by the orchestrator; hands back to the orchestrator on failure. Use when reading any upstream artefact handed off to VIGIL that carries a sibling .envelope.json.

Phase V (Verify) — establishes a reproducible failure. Deterministic-first; on two failed deterministic attempts, switches to statistical replay with confidence bands. No attribution begins without this phase passing. Use when starting any VIGIL mission, on mission intake, before any isolation or intervention work.

Load at the start of the Vivi Analyze phase. Techniques for building a repo map, running progressive disclosure, and maximising context relevance for coding tasks. Use whenever you need to understand unfamiliar code before editing it, or when an existing Discovery Report feels under-grounded.

Load when the Vivi Verify phase produces failures — test fails, lint error, build break, or three consecutive unsuccessful attempts at the same category. Provides the failure classification taxonomy, bounded debugging protocol, and escalation thresholds. Use to avoid the 'random retries' anti-pattern.

Load in Vivi's V (Verify) phase. The core capability that distinguishes Vivi from its predecessor APIVR-Δ: drive the closed, autonomous, bounded edit-run-test loop (`eidolons sandbox loop`) as its `--fix-hook`, consuming localized feedback, retrying in fresh context, refusing to game the tests, and gating on pass^k. Use whenever the task has an executable test oracle and a loop-competent host.

Load at Vivi session start, session end, or when you notice a repeating pattern across tasks. CRYSTALIUM-primary memory protocol: when mcp__crystalium__* tools are available, route ALL memory through CRYSTALIUM (recall/commit/ingest/plan_checkpoint/session_end). The local agents/memories/*.md files are the standalone fallback for when CRYSTALIUM is absent — EIIS conformance. Never write both.

Full Vivi cycle reference — A(nalyze) → P(lan) → I(mplement) → V(erify) → Δ(Delta)/R(eflect). Use when the task is a non-trivial feature implementation in a brownfield codebase and you need the complete methodology reference: complexity routing, evidence-grounded planning rules, test-anchoring requirements, and failure-escalation thresholds.

Load ONLY under TRANCE authorization when the Plan phase yields N independent, non-overlapping implementation tracks (disjoint file sets). Operationalizes the TRANCE G4 form — parallel multi-track implementation in mandatory git-worktree isolation, max 5 clean-context tracks, a per-track verifier cascade, a non-fungible per-track ≤3 reflection budget, explicit stop conditions, and a single-threaded merge/aggregation step. Absent TRANCE gating, Vivi runs exactly as today: single-track A→P→I→V→Δ/R.

Load when reading any upstream artefact handed off to Vivi that carries a sibling .envelope.json. BLOCKING per ECL §6.2.2: the orchestrator MUST have verified the envelope's SHA-256 (eidolons verify-envelope --block / eidolons run --verify) and recorded a verify_pass before dispatch. If no verify_pass exists for the message_id, or a verify_fail is present, REFUSE to process the payload and hand back to the orchestrator. Symmetric receiver gate — every Eidolon enforces it identically.