MainFrame vNext: publish curated lifecycle reference implementation

.agents/skills/classify-note.md → .agents/skills/classify-note/SKILL.md

@@ -32,11 +32,11 @@ Implemented (2026-06-13) as a focused sub-skill of ingest-source. Use it when th
 - Respect sensitivity overrides from `.context/routing-policy.md` (finance state etc. → Tier C, propose 20_live or park).
 
 ## Related
-- [ingest-source](ingest-source.md) — calls this (or equivalent logic) as step 3.
-- [extract-metadata](extract-metadata.md) — sibling for other frontmatter fields.
+- [ingest-source](../ingest-source/SKILL.md) — calls this (or equivalent logic) as step 3.
+- [extract-metadata](../extract-metadata/SKILL.md) — sibling for other frontmatter fields.
 - `.context/routing-policy.md` and `10_knowledge/index.md` — the source of truth for rules and domain inventory.
 
 ## Related
 
-- [ingest-source](ingest-source.md) — the parent skill
-- [extract-metadata](extract-metadata.md) — fills in the rest of frontmatter
+- [ingest-source](../ingest-source/SKILL.md) — the parent skill
+- [extract-metadata](../extract-metadata/SKILL.md) — fills in the rest of frontmatter

.agents/skills/craft-research-loop/SKILL.md

@@ -67,4 +67,4 @@ bin/craft-research-loop promote --project <slug> --trial <id> --to knowledge|exp
 - `.context/workflows/craft-research-loop.md`
 - `.context/templates/craft-trial.md`
 - research-lane-loop, project-experiment-loop
-- Reference implementation: `30_projects/image-generation-lab/outputs/RESEARCH_PROOF_INDEX.md`
+- Reference implementation: `local-only: 30_projects/<slug>/outputs/RESEARCH_PROOF_INDEX.md`

.claude/skills/create-source-summary.md → .agents/skills/create-source-summary/SKILL.md

@@ -8,7 +8,7 @@ status: active
 
 ## Purpose
 
-Some raw items (PDFs, long transcripts, scraped articles) are too dense for retrieval as-is. The ingest minion already creates a markdown stub for PDFs (per [ADR-007](../../DECISIONS.md)); this skill is the agent-driven version for markdown raw items that warrant a synthesized companion note.
+Some raw items (PDFs, long transcripts, scraped articles) are too dense for retrieval as-is. The ingest minion already creates a markdown stub for PDFs (per [ADR-007](../../../DECISIONS.md)); this skill is the agent-driven version for markdown raw items that warrant a synthesized companion note.
 
 The original raw file is preserved as evidence. A new `note`-type file is created alongside it, containing:
 - Summary of the source
@@ -28,5 +28,5 @@ Do not attempt to execute the extraction logic manually within this skill; the e
 
 ## Related
 
-- [ingest-source](ingest-source.md) — the parent skill
-- [ADR-007](../../DECISIONS.md) — defines the existing PDF stub-creation behavior
+- [ingest-source](../ingest-source/SKILL.md) — the parent skill
+- [ADR-007](../../../DECISIONS.md) — defines the existing PDF stub-creation behavior

.claude/skills/extract-metadata.md → .agents/skills/extract-metadata/SKILL.md

@@ -8,7 +8,7 @@ status: stub
 
 ## Purpose
 
-After [classify-note](classify-note.md) has set `domain`, `type`, and `tags`, fill in the rest of the frontmatter:
+After [classify-note](../classify-note/SKILL.md) has set `domain`, `type`, and `tags`, fill in the rest of the frontmatter:
 - **`source`** — preserve if already set; otherwise infer from filename, body, or capture context.
 - **`links`** — preserve the minion-extracted array; add agent-proposed connections.
 - **`status`** — transition `skimmed → routed` when starting work, `routed → extracted` when done.
@@ -22,5 +22,5 @@ After [classify-note](classify-note.md) has set `domain`, `type`, and `tags`, fi
 
 ## Related
 
-- [ingest-source](ingest-source.md) — the parent skill
-- [.context/primitives.md](../../.context/primitives.md) — schema and status lifecycle
+- [ingest-source](../ingest-source/SKILL.md) — the parent skill
+- [.context/primitives.md](../../../.context/primitives.md) — schema and status lifecycle

.agents/skills/ingest-source.md → .agents/skills/ingest-source/SKILL.md

@@ -10,7 +10,7 @@ status: implemented
 
 The canonical, reusable skill for the per-file judgment pass in the two-pass ingest architecture (ADR-009, ADR-011, ADR-019). It encapsulates everything between the deterministic minion pass 1 (which produces `status: skimmed` files in `01_ingest/ready/`) and the handoff to `bin/prep-ingest` (which moves to `queue/`) + minion pass 2 routing.
 
-The skill is invoked by the [ingest-agent](../../agents/ingest-agent.md) subagent. After this skill completes its proposal + user confirmation + enrichment, the file is ready for the strict deterministic gate.
+The skill is invoked by the [ingest-agent](../../../agents/ingest-agent.md) subagent. After this skill completes its proposal + user confirmation + enrichment, the file is ready for the strict deterministic gate.
 
 ## Inputs
 - A Markdown file in `01_ingest/ready/` with at minimum the minion-normalized frontmatter (`title`, `domain` possibly empty, `type`, `status: skimmed`, `source`, `tags`, and the `links:` array extracted from body wikilinks).
@@ -94,7 +94,7 @@ The skill is invoked by the [ingest-agent](../../agents/ingest-agent.md) subagen
 - `bin/prep-ingest`, `bin/ingest-minion`, `.context/workflows/audit-sweep.md`, `10_knowledge/index.md`, `.context/routing-policy.md`, `01_ingest/AGENTS.md`.
 
 ## Evaluation
-This skill should be exercised and measured during process evaluations (see `30_projects/mainframe-process-eval/`). Track: proposal acceptance rate, time-to-extracted, downstream routing success rate, number of needs-audit tags correctly applied, and any Tier B exceptions that later graduate to rules.
+This skill should be exercised and measured during process evaluations (see `40_operations/mainframe-process-eval/`). Track: proposal acceptance rate, time-to-extracted, downstream routing success rate, number of needs-audit tags correctly applied, and any Tier B exceptions that later graduate to rules.
 
 ## Status Note
 This skill was promoted from stub (2026-06-13) as part of closing the gap between the detailed agent procedure and reusable skill contracts. The long-form procedure now lives here; the ingest-agent definition should remain focused on role, guardrails, batch vs. per-file mode, and orchestration.

.agents/skills/mindgraph-retrieval/SKILL.md

@@ -1,6 +1,7 @@
 ---
 name: mindgraph-retrieval
-description: Use when an agent needs to retrieve durable knowledge or active project context from MainFrame's local graph-augmented search engine (MindGraph). It provides instructions on CLI/MCP commands and prevents hallucination of non-existent search verbs. Triggers on: "query mindgraph", "search mindgraph", "mindgraph query", "retrieve knowledge", "find in knowledge base", "query knowledge", "query projects", "mindgraph-refresh", "mindgraph doctor", "mindgraph status".
+description: >-
+  Use when an agent needs to retrieve durable knowledge or active project context from MainFrame's local graph-augmented search engine (MindGraph). It provides instructions on CLI/MCP commands and prevents hallucination of non-existent search verbs. Triggers on: "query mindgraph", "search mindgraph", "mindgraph query", "retrieve knowledge", "find in knowledge base", "query knowledge", "query projects", "mindgraph-refresh", "mindgraph doctor", "mindgraph status".
 status: active
 ---