fix(agent-context): deploy-all references in the C# + Python assemblers (cross-port parity)

The TS assembler was updated to deploy-all (emit every references/*.md regardless
of stack); the C# and Python ports still had the per-stack token filter, so the
byte-identity conformance gate failed for those ports.

Changes:
- server/python: drop `p.stem in stack.tokens` guard in assemble(); emit every
  .md in references/ sorted, matching TS deploy-all semantics.
- server/csharp: drop `.Where(stack.Tokens.Contains)` from the LINQ chain in
  AgentContextAssembler.Assemble(); same effect.
- fixtures/: regenerate conformance goldens via regen-agent-context-conformance.ts
  to pick up the 03bd9b5b doc additions (typescript.md +4 lines, migration.md
  +37 lines) that were missed when the goldens were last committed.

All conformance gates pass: Python 305/305, C# 670/670, CLI 35/35.

Author: dmealing

fixtures/agent-context-conformance/java-kotlin-react-tanstack/expected/.claude/skills/metaobjects-codegen/references/typescript.md

@@ -129,6 +129,10 @@ Generated files carry an `@generated by @metaobjectsdev/codegen-ts` header; the
 runner overwrites those and refuses to touch files without it. Hand-customizations
 that metadata can't express live in sibling `<Entity>.extra.ts` files.
 
+**Output format:** `meta gen` (and the CLI generally) is TTY-aware — human-readable
+text on a terminal, TOON on a pipe or agent. Override with `--format toon|json|text`.
+TOON is the structured default for agents; `--format json` is also available.
+
 ## Multiple output targets
 
 A `targets: { web: { outDir }, api: { outDir } }` registry plus a per-generator

fixtures/agent-context-conformance/java-kotlin-react-tanstack/expected/.claude/skills/metaobjects-verify/references/migration.md

@@ -21,8 +21,45 @@ npm install --save-dev @metaobjectsdev/cli @metaobjectsdev/migrate-ts
 You point the tool at the **same database your server connects to** — its
 connection is independent of your runtime tier.
 
+## Output format
+
+`meta migrate` (and the CLI generally) is TTY-aware: when stdout is a terminal it
+emits human-readable text; when piped to an agent or CI system it defaults to TOON
+(a compact, unambiguous machine-readable format). Override with `--format`:
+
+```bash
+meta migrate ... --format toon   # TOON (machine-readable, the pipe/agent default)
+meta migrate ... --format json   # JSON
+meta migrate ... --format text   # human-readable text (the TTY default)
+```
+
+Structured errors and next-step hints are also emitted on stdout (not stderr) in the
+active format, so callers can parse them without scraping stderr.
+
 ## The workflow
 
+### Fresh database: baseline first
+
+The default `meta migrate` path is **offline** — it diffs metadata against a
+committed schema snapshot rather than the live DB. On a fresh database there is no
+snapshot yet; run the `baseline` step once before the first migration generate:
+
+```bash
+meta migrate baseline --dialect sqlite     # seed snapshot from metadata (no DB needed)
+meta migrate baseline --dialect postgres   # same for Postgres
+meta migrate baseline --from-db --db postgresql://... --dialect postgres
+                                           # alternative: seed from live DB (for existing schemas)
+```
+
+`baseline` writes a reference snapshot to `.metaobjects/migrations/` and exits
+without emitting any SQL. After this, `meta migrate --dialect <d> --slug <name>`
+operates offline against that snapshot.
+
+If you run `meta migrate` before baselining, the CLI surfaces a structured
+next-step hint pointing to the exact `baseline` command.
+
+### Generating a migration
+
 1. **Generate a migration** by diffing metadata vs the prior state (the live DB or a
    committed snapshot). The engine emits paired `up.sql` + `down.sql`:
 

fixtures/agent-context-conformance/java-react/expected/.claude/skills/metaobjects-codegen/references/typescript.md

@@ -129,6 +129,10 @@ Generated files carry an `@generated by @metaobjectsdev/codegen-ts` header; the
 runner overwrites those and refuses to touch files without it. Hand-customizations
 that metadata can't express live in sibling `<Entity>.extra.ts` files.
 
+**Output format:** `meta gen` (and the CLI generally) is TTY-aware — human-readable
+text on a terminal, TOON on a pipe or agent. Override with `--format toon|json|text`.
+TOON is the structured default for agents; `--format json` is also available.
+
 ## Multiple output targets
 
 A `targets: { web: { outDir }, api: { outDir } }` registry plus a per-generator

fixtures/agent-context-conformance/java-react/expected/.claude/skills/metaobjects-verify/references/migration.md

@@ -21,8 +21,45 @@ npm install --save-dev @metaobjectsdev/cli @metaobjectsdev/migrate-ts
 You point the tool at the **same database your server connects to** — its
 connection is independent of your runtime tier.
 
+## Output format
+
+`meta migrate` (and the CLI generally) is TTY-aware: when stdout is a terminal it
+emits human-readable text; when piped to an agent or CI system it defaults to TOON
+(a compact, unambiguous machine-readable format). Override with `--format`:
+
+```bash
+meta migrate ... --format toon   # TOON (machine-readable, the pipe/agent default)
+meta migrate ... --format json   # JSON
+meta migrate ... --format text   # human-readable text (the TTY default)
+```
+
+Structured errors and next-step hints are also emitted on stdout (not stderr) in the
+active format, so callers can parse them without scraping stderr.
+
 ## The workflow
 
+### Fresh database: baseline first
+
+The default `meta migrate` path is **offline** — it diffs metadata against a
+committed schema snapshot rather than the live DB. On a fresh database there is no
+snapshot yet; run the `baseline` step once before the first migration generate:
+
+```bash
+meta migrate baseline --dialect sqlite     # seed snapshot from metadata (no DB needed)
+meta migrate baseline --dialect postgres   # same for Postgres
+meta migrate baseline --from-db --db postgresql://... --dialect postgres
+                                           # alternative: seed from live DB (for existing schemas)
+```
+
+`baseline` writes a reference snapshot to `.metaobjects/migrations/` and exits
+without emitting any SQL. After this, `meta migrate --dialect <d> --slug <name>`
+operates offline against that snapshot.
+
+If you run `meta migrate` before baselining, the CLI surfaces a structured
+next-step hint pointing to the exact `baseline` command.
+
+### Generating a migration
+
 1. **Generate a migration** by diffing metadata vs the prior state (the live DB or a
    committed snapshot). The engine emits paired `up.sql` + `down.sql`:
 

fixtures/agent-context-conformance/python/expected/.claude/skills/metaobjects-codegen/references/typescript.md

@@ -129,6 +129,10 @@ Generated files carry an `@generated by @metaobjectsdev/codegen-ts` header; the
 runner overwrites those and refuses to touch files without it. Hand-customizations
 that metadata can't express live in sibling `<Entity>.extra.ts` files.
 
+**Output format:** `meta gen` (and the CLI generally) is TTY-aware — human-readable
+text on a terminal, TOON on a pipe or agent. Override with `--format toon|json|text`.
+TOON is the structured default for agents; `--format json` is also available.
+
 ## Multiple output targets
 
 A `targets: { web: { outDir }, api: { outDir } }` registry plus a per-generator

fixtures/agent-context-conformance/python/expected/.claude/skills/metaobjects-verify/references/migration.md

@@ -21,8 +21,45 @@ npm install --save-dev @metaobjectsdev/cli @metaobjectsdev/migrate-ts
 You point the tool at the **same database your server connects to** — its
 connection is independent of your runtime tier.
 
+## Output format
+
+`meta migrate` (and the CLI generally) is TTY-aware: when stdout is a terminal it
+emits human-readable text; when piped to an agent or CI system it defaults to TOON
+(a compact, unambiguous machine-readable format). Override with `--format`:
+
+```bash
+meta migrate ... --format toon   # TOON (machine-readable, the pipe/agent default)
+meta migrate ... --format json   # JSON
+meta migrate ... --format text   # human-readable text (the TTY default)
+```
+
+Structured errors and next-step hints are also emitted on stdout (not stderr) in the
+active format, so callers can parse them without scraping stderr.
+
 ## The workflow
 
+### Fresh database: baseline first
+
+The default `meta migrate` path is **offline** — it diffs metadata against a
+committed schema snapshot rather than the live DB. On a fresh database there is no
+snapshot yet; run the `baseline` step once before the first migration generate:
+
+```bash
+meta migrate baseline --dialect sqlite     # seed snapshot from metadata (no DB needed)
+meta migrate baseline --dialect postgres   # same for Postgres
+meta migrate baseline --from-db --db postgresql://... --dialect postgres
+                                           # alternative: seed from live DB (for existing schemas)
+```
+
+`baseline` writes a reference snapshot to `.metaobjects/migrations/` and exits
+without emitting any SQL. After this, `meta migrate --dialect <d> --slug <name>`
+operates offline against that snapshot.
+
+If you run `meta migrate` before baselining, the CLI surfaces a structured
+next-step hint pointing to the exact `baseline` command.
+
+### Generating a migration
+
 1. **Generate a migration** by diffing metadata vs the prior state (the live DB or a
    committed snapshot). The engine emits paired `up.sql` + `down.sql`:
 

fixtures/agent-context-conformance/ts-react-tanstack/expected/.claude/skills/metaobjects-codegen/references/typescript.md

@@ -129,6 +129,10 @@ Generated files carry an `@generated by @metaobjectsdev/codegen-ts` header; the
 runner overwrites those and refuses to touch files without it. Hand-customizations
 that metadata can't express live in sibling `<Entity>.extra.ts` files.
 
+**Output format:** `meta gen` (and the CLI generally) is TTY-aware — human-readable
+text on a terminal, TOON on a pipe or agent. Override with `--format toon|json|text`.
+TOON is the structured default for agents; `--format json` is also available.
+
 ## Multiple output targets
 
 A `targets: { web: { outDir }, api: { outDir } }` registry plus a per-generator

fixtures/agent-context-conformance/ts-react-tanstack/expected/.claude/skills/metaobjects-verify/references/migration.md

@@ -21,8 +21,45 @@ npm install --save-dev @metaobjectsdev/cli @metaobjectsdev/migrate-ts
 You point the tool at the **same database your server connects to** — its
 connection is independent of your runtime tier.
 
+## Output format
+
+`meta migrate` (and the CLI generally) is TTY-aware: when stdout is a terminal it
+emits human-readable text; when piped to an agent or CI system it defaults to TOON
+(a compact, unambiguous machine-readable format). Override with `--format`:
+
+```bash
+meta migrate ... --format toon   # TOON (machine-readable, the pipe/agent default)
+meta migrate ... --format json   # JSON
+meta migrate ... --format text   # human-readable text (the TTY default)
+```
+
+Structured errors and next-step hints are also emitted on stdout (not stderr) in the
+active format, so callers can parse them without scraping stderr.
+
 ## The workflow
 
+### Fresh database: baseline first
+
+The default `meta migrate` path is **offline** — it diffs metadata against a
+committed schema snapshot rather than the live DB. On a fresh database there is no
+snapshot yet; run the `baseline` step once before the first migration generate:
+
+```bash
+meta migrate baseline --dialect sqlite     # seed snapshot from metadata (no DB needed)
+meta migrate baseline --dialect postgres   # same for Postgres
+meta migrate baseline --from-db --db postgresql://... --dialect postgres
+                                           # alternative: seed from live DB (for existing schemas)
+```
+
+`baseline` writes a reference snapshot to `.metaobjects/migrations/` and exits
+without emitting any SQL. After this, `meta migrate --dialect <d> --slug <name>`
+operates offline against that snapshot.
+
+If you run `meta migrate` before baselining, the CLI surfaces a structured
+next-step hint pointing to the exact `baseline` command.
+
+### Generating a migration
+
 1. **Generate a migration** by diffing metadata vs the prior state (the live DB or a
    committed snapshot). The engine emits paired `up.sql` + `down.sql`:
 

server/csharp/MetaObjects/AgentContext/AgentContextAssembler.cs

@@ -50,7 +50,8 @@ public static IReadOnlyList<AssembledFile> Assemble(string contentRoot, Stack st
         output.Add(new AssembledFile(".metaobjects/AGENTS.md", alwaysOn));
         output.Add(new AssembledFile(".metaobjects/CLAUDE.md", alwaysOn));
 
-        // 2. Skills: body + only the references whose token is in the stack.
+        // 2. Skills: body + ALL references (deploy-all: every fragment is installed
+        //    regardless of stack; robust to stack-detection misses).
         foreach (var skill in Stack.SkillNames)
         {
             var skillDir = Path.Combine(contentRoot, "skills", skill);
@@ -60,12 +61,11 @@ public static IReadOnlyList<AssembledFile> Assemble(string contentRoot, Stack st
             var refDir = Path.Combine(skillDir, "references");
             if (Directory.Exists(refDir))
             {
-                // Tokens (filename-without-".md") in the stack, sorted ascending (ordinal).
+                // All tokens (filename-without-".md"), sorted ascending (ordinal).
                 var tokens = Directory.EnumerateFiles(refDir)
                     .Select(Path.GetFileName)
                     .Where(n => n is not null && n.EndsWith(".md", StringComparison.Ordinal))
                     .Select(n => n![..^".md".Length])
-                    .Where(stack.Tokens.Contains)
                     .OrderBy(t => t, StringComparer.Ordinal)
                     .ToArray();
                 foreach (var token in tokens)

server/python/src/metaobjects/agent_context/assemble.py

@@ -107,7 +107,8 @@ def assemble(content_root: Path, stack: Stack) -> list[AssembledFile]:
     out.append(AssembledFile(".metaobjects/AGENTS.md", always_on))
     out.append(AssembledFile(".metaobjects/CLAUDE.md", always_on))
 
-    # 2. Skills: body + only the references whose token is in the stack.
+    # 2. Skills: body + ALL references (deploy-all: every fragment is installed
+    #    regardless of stack; robust to stack-detection misses).
     for skill in SKILL_NAMES:
         skill_dir = content_root / "skills" / skill
         body = _read_text(skill_dir / "SKILL.md")
@@ -118,7 +119,7 @@ def assemble(content_root: Path, stack: Stack) -> list[AssembledFile]:
             tokens = sorted(
                 p.stem
                 for p in ref_dir.iterdir()
-                if p.is_file() and p.suffix == ".md" and p.stem in stack.tokens
+                if p.is_file() and p.suffix == ".md"
             )
             for token in tokens:
                 out.append(