Skip to content

Add the write tools for change sets and content types - #2514

Merged
pournasserian merged 6 commits into
v2from
2509-write-tools
Oct 18, 2026
Merged

pournasserian merged 6 commits into
v2from
2509-write-tools

Conversation

@pournasserian

Copy link
Copy Markdown
Member

Closes #2509

What

The write tools of F-MCP (#2423) for content types and change sets, on the conventions from #2508. An agent proposes schema changes and submits them, and a person reviews and publishes them.

  • Change sets: open_change_set (rationale, sources), submit_change_set (open to checks running) and reopen_change_set (checks failed to open), for the agent's own sets only (Refuse agents' review moves and enforce change set caps #2503).
  • Content types: create_content_type (name, slug, definition) and revise_content_type (slug, base version, definition).
    • Each stores the new version, unpublished (Add atomic change set publish #2482), and adds the matching schema operation to a change set in one unit of work, so either both are written or neither is.
    • The set is a change_set_id from open_change_set, or a session_id with a rationale and sources, which opens the session's set on its first call (CHG-02).
  • Each write result carries the content type, the change set and a next sentence that says what comes next.
  • Core:
    • IChangeSetDirectory.GetSessionSetAsync reads the caller's open set for a session id.
    • IAgentPermissions.CheckChangeSetSizeAsync takes the number of operations a write is about to add.
    • ToolArgumentsReader reads lists of strings, objects, optional ids and required whole numbers.
  • CONTRIBUTING.md: the read and write tools and what is live when.

Why

MCP-01 asks for tools for every capability, and the governance model needs everything an agent writes to wait in a change set for a person. With #2482, a stored content type version is not published until its set is. With this PR, every version an agent stores is also in one of its change sets, and no tool approves, rejects or publishes (MCP-06).

Choices worth a look:

  • One unit of work for both writes. Each directory finds its expected failures before its own unit of work, so the tool's outer one undoes the version when the change set refuses the operation, for example because of the item's version, a cap or ownership.
  • Caps are checked before the write too. A refused cap inside the outer unit of work would roll back its own security event (SEC-08, the limit noted in Refuse agents' review moves and enforce change set caps #2503). The tool therefore first reads the target set and asks IAgentPermissions with the operation it is about to add, or asks for a new set when the session has none. That refusal commits its event, and nothing is stored. The set checks again on the add, which settles a race.
  • GetSessionSetAsync is a small read on the change set directory, the one place that knows a session's open set. The pre-check needs it.
  • No operation scope yet: an agent with any key can propose. Type, field and operation scopes are the credentials lane's.
  • Create and revise of one type in one set is refused by the set (one operation per item), with the set's own message.
  • Submit and reopen are separate tools, so each description says exactly when to call it.
  • Not in this PR: approve, reject, publish and revert (never for agents), bulk operations with progress and the MCP-01 coverage test (next slice), and entry and entity writes (their lanes).

Tests

Integration, over HTTP:

  • On SQLite and PostgreSQL, an agent opens a set, creates category, revises a person's product from version 1 and submits the set. The set holds both operations in order, is in checks running and names the agent, and the tool call log has all four calls, succeeded.
  • On SQLite and PostgreSQL, two creates with one session_id land in one set with the first call's rationale.
  • On SQLite and PostgreSQL, these store no version, while the agent's own valid create is stored:
    • a definition with an invalid slug, an unknown field type and a bad field name returns every error at once
    • a create past a cap of one item is refused, and agent.items_per_change_set_exceeded stays in the log
    • a create into a person's set and a submit of it are refused as not the agent's own
    • a create without a target names the missing argument
  • A revision from a stale base version returns version_conflict with a hint to read the current version, and the type stays at version 2.
  • The conformance tests now cover the write tools: names, described schemas with valid examples, and unknown arguments refused before any service is read. Unit tests cover the new reader methods.

The full suite ran locally against a postgres:17 container as well as SQLite, and markdownlint passes. No new package and no new project reference.

Checklist

  • Linked to an issue a maintainer has confirmed
  • Targets v2
  • Tests or docs updated where behaviour changed
  • No tool-generated co-author trailers or footers

@pournasserian pournasserian added this to the M1 Governance core milestone Oct 18, 2026
@pournasserian pournasserian added type:feature New capability v2 FluentCMS v2 (v2 branch) area:mcp MCP server and agent credentials (MCP) labels Oct 18, 2026
@pournasserian
pournasserian merged commit 64f5421 into v2 Oct 18, 2026
2 checks passed
@pournasserian
pournasserian deleted the 2509-write-tools branch October 18, 2026 23:29
pournasserian added a commit that referenced this pull request Oct 18, 2026
Closes #2515

## What

The next F-CNT (#2416) slice: change sets hold operations on entries and
entities, and diff, publish and revert them like schema operations
(CHG-01, CHG-03, CHG-05, CHG-06). Builds on atomic publish (#2482) and
revert (#2511). This is the first thing that moves
`Entries.PublishedVersion`.

- **Item kinds**: an `entry` operation must point at an entry of a plain
content type, and an `entity` operation at an entry of an entity type,
in each case of a type the agent's scope allows for the operation
(MCP-05). The wrong kind fails with the new `item_kind_mismatch`. A
missing or soft-deleted entry fails with `item_not_found`, and a missing
version with `version_not_found`. `change_set.operation_added` carries
the base version's row as previous version, as for schema items. Other
kinds (media, rules) are still `item_kind_unavailable`.
- **Diff** (`ItemDiff.ForEntry` in Core): two versions of an entry,
field by field, with `ValueDiff.Compare`. On an entity a field's value
is its whole fact, so a new source shows as a change. Fields follow the
definition the shown version was written against, so the order is the
same on SQLite and PostgreSQL even though `jsonb` reorders keys.
`DiffAsync` reads every version the entry operations name, with their
definitions, in one query.
- **Publish**: each entry's published version moves from its operation's
base version to the proposed one with a conditional update, alongside
schema items in the same unit of work. Stale bases fail with
`base_version_stale` before anything moves, and a soft-deleted entry
fails with `item_not_found`. Events: `entry.published` /
`entry.unpublished`, or `entity.*`.
- **Delete means unpublish**: a published delete clears the published
version and keeps the entry, its draft and its history. Revert of a
create unpublishes, and revert of a delete publishes the base version
again, through the same publish.
- **One history per entity**: `EntryDirectory` now names entries of
entity types `entity` in its events (`entity.created`, `entity.updated`
and so on), so an entity's own events and the change set events about it
share a subject.

## Why

Entries were stored and versioned (#2491) but nothing could put them
live. Publishing through change sets keeps CHG-05's all-or-nothing rule
for mixed sets of schema, entry and entity items (CHG-09), and revert
comes for free from `ChangeOperation.Inverse`.

Choices worth a look:

- **Separate kinds for entries and entities** let the desk and blast
radius (ENT-04) tell content changes from fact changes, without a lookup
per item.
- **A published delete unpublishes rather than soft-deletes**, so its
revert only has to move a pointer back. Soft delete stays a direct
action with its retention window (#2502). The entry states slice will
call an unpublished entry archived.
- **Writes do not join change sets themselves yet.** A write stores a
draft version and an operation points at it, as content types did before
SCH-05. Entry writes will follow the change set target that #2514
introduced, together with the entry tools.
- **Changes to the F-CHG lane's file are kept in separate methods**
(`ResolveEntryAsync`, `EntryDiffsAsync`, `PublishedEntriesAsync`,
`PublishableEntry`), so the SCH-05 work merges easily.

## Tests

Unit (Core):
- `ItemDiff.ForEntry`: changed fields in definition order, with a
removed field the type no longer has last; formatting-only differences
ignored; facts compared whole; create and delete; refusal of non-entry
operations and of fields that do not match the operation's versions.

Integration, each on SQLite and PostgreSQL:
- An agent cannot add an operation on an entry of a type outside its
scope.
- Entry and entity operations stored with their kinds and events; the
wrong kind both ways, a missing entry, a soft-deleted entry and a
missing version each fail and write nothing; an update's event carries
its base version's row.
- The diff of an entry update and an entity create.
- Publishing moves both pointers with one event per item; a resolved
view follows the published price, not a new draft; an entity's own and
change set events share its history.
- A stale base and a soft-deleted entry fail the publish and move
nothing.
- A published delete unpublishes and keeps the entry; its revert
publishes it again and marks the delete reverted.
- Reverting an update puts the earlier version back, and its diff shows
the change backwards.
- A failure injected at each audit append of a publish leaves no entry
published.

The full suite ran locally against a `postgres:17` container as well as
SQLite, after merging the latest v2. No migration, no new package and no
new project reference.

## Checklist

- [x] Linked to an issue a maintainer has confirmed
- [x] Targets `v2`
- [x] Tests or docs updated where behaviour changed
- [x] No tool-generated co-author trailers or footers
This was referenced Oct 19, 2026
pournasserian added a commit that referenced this pull request Oct 19, 2026
Closes #2523

## What

The last F-CNT (#2416) slice: the entry tools agents call over MCP
(MCP-01, P1), following the write-tool conventions of #2514.

- **`EntryTools`** (`FluentCMS.Core.Builtin.Tools`), discovered like the
other built-in providers:
- Reads: `list_entries`, `get_entry` (latest draft or a named version,
with state and pointers), `get_entry_history`, `get_resolved_entry`
(entity-field references replaced by published values, plus what could
not be resolved) and `get_where_used` (an entry, one entity field with
`field`, or a media asset with `asset_id`).
- Proposals: `create_entry`, `update_entry` and `delete_entry`. The new
version, if any, and its operation commit together into a set named by
`change_set_id` or a session's set (`ChangeSetTarget`), as
`create_content_type` does, so a refused add stores no version. The
operation is an entity operation for an entity type.
- **What an update proposes**: it is based on the entry's published
version, the version publishing replaces. An entry that was never
published gets a create of the new version, so publishing it never fails
as stale just because drafts piled up. `delete_entry` proposes
unpublishing, and an entry that is not published is refused with the new
`entry_not_published`.
- **Hints** name the next call: `get_entry` after a version conflict,
`get_schema_document` for invalid fields, `list_entries` for a missing
entry or a bad reference, and `list_content_types` for an unknown type.
- `ToolJson` renders entries, summaries, history, resolved views and
usages; states and reasons are snake_case like the rest.
- Tests: the CNT row of `McpCoverageTests` lists the tools,
`EntryToolsConformanceTests` covers the provider, and the MCP test
session (`ConnectedAgent`) moved out of `WriteToolsTests` so both test
classes share it, with a narrower scope where a test needs one.

## Why

P1 has every capability as a tool before it is a screen. These are the
entry half of the reference scenario: an agent writes a product and a
page that shows its price, and a person reviews and publishes them.

Choices worth a look:

- **The propose helper is a copy of `ProposedWrite`'s shape**, kept
inside `EntryTools` and not merged with the schema one, because SCH-05
is about to move the change set target into the directories. Entries
follow that change when it lands.
- **Soft delete and restore stay off the tool list**: they bypass change
sets, so they remain a person's action for now.
- **A second update of one entry in one set is refused as a duplicate
operation**, as a second revision of a content type is today. Amending
an operation in place is change set work.
- **Entity-specific tools** (facts, validity, blast radius) are left to
F-ENT; the ENT row of the coverage test stays open.

## Tests

Unit (Core):
- `EntryToolsConformanceTests`: names, descriptions, input schemas with
examples, unknown arguments refused.
- `McpCoverageTests`: the new tools are registered, listed, and none
approves or publishes.

Integration over MCP, each on SQLite and PostgreSQL:
- S2 end to end: an agent creates a product entity and a page that
references its price in one set and submits it. The page is in review
until a person publishes the set. Then the agent reads the page resolved
with the price, finds the page where `product.price` is used, and lists,
reads and pages through the history.
- An update of a published entity proposes an update from its published
version; a stale base version is refused with a hint to read the entry,
invalid fields with a hint to read the schema document, and neither
stores a version or opens a set.
- `delete_entry` refuses an unpublished entry and proposes unpublishing
a published one.
- An agent scoped to products cannot read or resolve a product page,
where used leaves the page out, and `get_where_used` without a target is
refused.

The full suite ran locally against a `postgres:17` container as well as
SQLite. No migration, no new package and no new project reference.

## Checklist

- [x] Linked to an issue a maintainer has confirmed
- [x] Targets `v2`
- [x] Tests or docs updated where behaviour changed
- [x] No tool-generated co-author trailers or footers
This was referenced Oct 19, 2026
pournasserian added a commit that referenced this pull request Oct 19, 2026
Closes #2534

## What

The F-ENT (#2417) entity tools (MCP-01, P1). Agents read facts,
relations, unsourced facts and the blast radius, and propose one fact
with its source, or a relation, in the write-tool conventions of #2514.

- **`EntityTools`** (`FluentCMS.Core.Builtin.Tools`):
- Reads: `get_entity_facts` (source, owner, validity and the unsourced
flag, for the latest draft or a version), `get_entity_relations` (both
directions), `list_unsourced_facts` (the site or one entity type) and
`get_blast_radius` (the published entries a set's entity changes affect,
with old and new values per path).
- `set_entity_fact` proposes one fact with its value, source, owner and
validity, and keeps the other facts. A property left out is stored as
null, so a fact without a source is flagged unsourced.
- `relate_entities` proposes a relation. It sets a single reference
field, or adds to a list of references without duplicates, and keeps the
field's metadata. A field that is not a reference field is refused with
the new `field_not_relation`. A plain entry is refused with
`entry_not_entity`.
- Both writes read the version given as `base_version`, change it in a
copy, and propose an update from the published version, or a create
while the entity was never published, as `update_entry` does.
- **`ProposedEntryWrite`**: the propose flow `EntryTools` used, moved
into a shared helper with the update rule and the hints, so entry and
entity writes commit a version and its operation the same way.
- **`ToolArgumentsReader`** gains `RequiredValue`, for an argument that
can be any JSON value but null (a fact's value), and `OptionalObject`.
- `ToolJson` renders facts, flags, relations and the blast radius. The
ENT rows of `McpCoverageTests` list the new tools; stale facts stay open
on #2417.

## Why

S1's pipeline agent changes one fact per product, with the export as its
source. With `update_entry` alone it would have to resend each whole
entity. `set_entity_fact` and `relate_entities` change only the fact or
relation they name, and go through the same validation, reference
checks, fact index and change set as any entry write. The reads make the
earlier F-ENT slices (#2528, #2530, #2533) reachable over MCP.

Choices worth a look:

- **Left-out properties are null, not kept**: a new value usually needs
a new source and validity, so keeping the old ones would make a stale
source look current. Leaving out a source makes the fact show as
unsourced.
- **A list relation is appended, never replaced**: removing one is an
`update_entry`.
- **Stale facts** come with ENT-03, which waits for the job queue
(#2531).

## Tests

Unit (Core):
- `EntityToolsConformanceTests`; `McpCoverageTests` with the new rows.

Integration over MCP, each on SQLite and PostgreSQL:
- S1: two prices set from the export in one session's set; the blast
radius lists exactly the two products' published pages with the old and
new price, not the third product's page. The fact reads back sourced.
- A fact set without a source reads back flagged and is listed by
`list_unsourced_facts`.
- `relate_entities` adds a spec and sets a category;
`get_entity_relations` shows both outgoing, and the spec shows the
product incoming.
- A base version that does not exist, a field the type lacks (with a
hint to the schema document), a plain entry, a field that is not a
reference (with a hint to the content type) and a related entity of the
wrong type are all refused, and nothing is stored.

The full suite ran locally against a `postgres:17` container as well as
SQLite. No migration, no new package and no new project reference.

## Checklist

- [x] Linked to an issue a maintainer has confirmed
- [x] Targets `v2`
- [x] Tests or docs updated where behaviour changed
- [x] No tool-generated co-author trailers or footers
pournasserian added a commit that referenced this pull request Oct 19, 2026
Closes #2513

## What

The SCH-05 slice of F-SCH (#2415), on top of change set publish (#2482)
and revert (#2511). Every content type write is now a change set
operation. A create or a revision stores its new version and records the
operation in the change set it names, in one unit of work. Nothing a
write does is live until that set publishes.

- **Core** (`FluentCMS.Core.ChangeSets`):
- `ChangeSetTarget`: the set a write joins, either an existing set by id
or a session id with the rationale and sources of the set the session's
first write opens (CHG-02).
- `ChangeSet.Record(operation)` adds the operation, or extends the one
the set already holds for the item. The write must be based on the
version that operation proposes, and the operation then proposes the
write's version. A create stays a create and an update keeps its base,
so a set still has one operation per item (CHG-03) and its diff runs
from the base to the newest version.
- **Core** (`FluentCMS.Core.Schema`):
- `IContentTypeDirectory.CreateAsync` and `ReviseAsync` take a
`ChangeSetTarget`. A create records a create proposing version 1, and a
revision records an update from its base to the new version.
- A revision's base is the latest version or the published one. In a set
that already holds the type it is the version that set proposes, so the
set can keep revising its own draft after another set drafts the type.
Basing it on the published version leaves an unpublished draft behind,
such as a rejected one. The referenced-field rule (SCH-04) then also
checks the fields of every draft an open set proposes, so such a
revision cannot drop a field another draft references.
- A revision of a type no set has published, or one a published delete
took offline (a rejected or reverted create), records a create, since
there is nothing live to update. `ChangeSet.Record` takes whether the
item is published, and an agent needs the create scope for such a
revision (MCP-05).
- A draft is a base only while a set that can still publish proposes it.
A revision of a rejected or reverted draft is refused at once with
`version_conflict` and points to the published version, instead of
failing later at publish. While an open set proposes a draft, the latest
or an older one, a revision keeps its referenced fields and its entity
flag (SCH-03). A draft whose set went stale still counts until a person
rejects it; #2588 tracks that.
- References between types are checked against what each type is heading
to: every draft an open set proposes, or its published version while
there is none. A rejected draft neither drops a live reference nor
offers a field to reference, and a field a pending draft removes is no
target. A field another type references stays while any of that type's
live versions or open drafts references it.
- `GetPublishedAsync(slug)` reads the live version, or fails with
`content_type_not_published`. `ContentTypeSummary` carries
`PublishedVersion`.
- **Infrastructure**:
- `ChangeSetDirectory` gains an internal join for writes owned by
another directory. `PrepareAsync` finds the expected failures before the
caller's unit of work begins: unknown set, set not open, blank
rationale, agent ownership and the items cap. Inside it,
`BeginWriteAsync` opens the set or moves its revision, and `RecordAsync`
records the operation after the item's own rows and event. An agent's
daily cap is counted in `BeginWriteAsync` with its credential held, as
`OpenAsync` does since #2580. It is therefore the one refusal that rolls
a caller's unit of work back, and #2513's acceptance now says so.
- `content_type.created` and `content_type.revised` now carry the set's
id, and come after `change_set.opened` when the write opens its
session's set. An extended operation appends
`change_set.operation_updated`, the name #2575 uses for the same move of
an entry operation.
  - Both directories share one `ChangeSetDirectory` per scope.
- **MCP write tools** (#2514): `create_content_type` and
`revise_content_type` pass their target to the directory, which now
records the operation itself. `ProposedWrite` no longer opens its own
unit of work or checks the set ahead of the write; the directory's
checks run before anything is written, so a refused cap still keeps its
security event. The answer names the kind the set recorded, so a
revision of a type with nothing live reads as a create. In
`WriteToolsTests` the agent's revision of the person's unpublished
product is now expected as `Create null→2` instead of `Update 1→2`,
which is the behaviour change agents see. The entry tools (#2525) keep
the tool-level pre-check and add, since entry writes do not join change
sets in their directory yet.
- No migration: both pointers already exist.

## Why

The decision, as written in #2513: `Version` is the latest version,
which authoring builds on, and `PublishedVersion` is what is live. The
authoring reads stay on the latest version: `GetAsync`, `ListAsync`, the
schema document and entry validation. A set can then build on its own
drafts, and entries can be written against the schema change they need
in the same set (CHG-09). Delivery reads the published version when
F-API arrives.

Choices worth a look:

- **The operation is written by the change set directory, inside the
content type directory's unit of work.** Version and operation commit
together, and the existing "expected failures before the unit of work"
rule still holds for both.
- **Extending instead of refusing a second write.** An agent iterating
on a type within a session would otherwise need a new set for every
revision.
- **A unique index at commit** means the slug or version race it always
meant. When the write opened a session's set, it may also be a
concurrent first write of that session, so that case answers with
`change_set_conflict`.
- **Two revisions racing into the same set**: the loser now fails with
`change_set_conflict`, since the set's revision moved first, rather than
`version_conflict`. The race tests give each writer its own set to keep
checking the version race.
- **Reference checks at write time are best effort.** Sets publish in
any order, and a revert can take a field away again, so a reference can
still dangle across sets. #2604 checks references at publish, which is
where the order is known. It also covers dropping a field together with
the last reference to it in one set, which the write-time check still
refuses.
- **Breaking signature**: every caller of `CreateAsync` and
`ReviseAsync` passes a target. In this PR that is the MCP schema tools
and tests: the schema, entry and MCP scenarios write their content model
in a `content-model` session set, as a person, so agent caps and the
change set reads in tests are not affected.
- **Answering from the set the write joined**: the schema tools read the
session's set again after the write, which can race with a submit; #2615
passes the set out of the write.
- **Not in this PR**: entries as change set operations, and refusing at
publish an entry whose content type version is not published (F-CNT).
Also left out: deleting a content type, the MCP schema tools, the desk's
schema review and delivery reading the published schema.

## Tests

Unit (Core):

- A later write extends the item's operation, keeping its kind and base,
for a create and an update.
- A later write not based on the proposed version, a delete over an
update, and an update over a delete are refused and change nothing.
- A write is recorded only while the set is open.

Integration, each on SQLite and PostgreSQL:

- A create is an operation of its set, with `content_type.created`
naming the set. The type lists with no published version and
`GetPublishedAsync` refuses it until the set publishes.
- A revision in the same set extends the operation (create from nothing
to version 2). Its events name the version it was based on, and the diff
lists both fields as added.
- A revision of a published type into a new set is an update from the
published version. It is live only after that set publishes, and the
previous version stays readable.
- After a rejected draft, a revision based on the published version goes
through, while one based on the rejected draft fails with
`version_conflict`.
- After a rejected create, and after a published create was reverted,
the next revision is recorded as a create, and publishing it makes the
type live again.
- A revision from the published version cannot drop a field that another
draft in an open set references.
- Writes through a session share the set the first one opens.
- An unknown set, a set that is not open and a blank session rationale
are refused inside a caller's unit of work, which then commits.
- Inside a caller's unit of work, an agent cannot write into another
agent's set, and a write past its items cap is refused. Neither stores a
content type or change set event, the cap refusal keeps its security
event, and the caller's unit of work commits.
- A failure injected at either audit append of a revision, after the
pointer and the set's revision reached the database, leaves neither the
version nor the operation. A caller's unit of work rolled back after the
revision's rows were saved leaves neither either.
- A rejected draft is refused as a base, while a draft an open set
proposes can be built on and publishes after it.
- An agent without the create scope cannot revise a type with nothing
live.
- A revision from the published version keeps an open draft's entity
flag, and may drop it once that draft's set is rejected.
- A rejected draft of a page does not let its brand drop a field the
live page references, and a field only a rejected draft has cannot be
referenced.
- A field that a pending draft removes cannot be referenced, from the
draft's own set or another one.
- A field that an older open draft references cannot be removed, even
when a later draft of that type is the latest.
- A set keeps revising its own draft after another set drafted the type,
and a revision of it from the latest version is refused with
`version_conflict`.
- A revision into a rejected set that holds the type is refused with
`change_set_not_open`.
- A revision keeps the entity flag of an older open draft when the
latest draft was rejected.
- A content type write that would open a set past an agent's daily cap
is refused, stores no set or type, and keeps one security event.
- A session's first write appends the set's opening before the content
type's own event.

`dotnet build -warnaserror` and the full suite pass locally on SQLite
and against a `postgres:17` container. No new package and no new project
reference.

## Checklist

- [x] Linked to an issue a maintainer has confirmed
- [x] Targets `v2`
- [x] Tests or docs updated where behaviour changed
- [x] No tool-generated co-author trailers or footers
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area:mcp MCP server and agent credentials (MCP) type:feature New capability v2 FluentCMS v2 (v2 branch)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Write tools: open and submit change sets, and create and revise content types into them

1 participant