Watch
0
0
Fork
You've already forked hyperhive
0

agents get read, not write, on their own agent-configs repo; an existing grant is left alone

This commit is contained in:
damocles 2026-11-15 16:20:54 +01:00 • committed by mara
commit 07c66e3a1c
7 changed files with 35 additions and 41 deletions

View file

@ -125,7 +125,7 @@ checkpoints**, not about sandboxing the agent from its own tools:
which swarm-controller converges on every config repo. An operator's merge
there is also what deploys the config. Existing `agents/<repo>` repos carry
the same merge gate.
An agent (a write collaborator, not a repo admin) can neither change those
An agent (a read collaborator, not a repo admin) can neither change those
settings nor merge its own PR. It's **not** set up for external VCS (GitHub
etc.), though — there, operator-merge is process + accepted risk, not a
technical control.

View file

@ -33,7 +33,7 @@ agent gets every new item, and two of them would both label and assign it.
creating a `triage` agent gives each new issue two todos. Switch one off on the
board: the custom flow, or the archetype's flow
([below](#on-the-board)).
- **Forge rights.** A new agent's forge account is a write collaborator on its own
- **Forge rights.** A new agent's forge account is a read collaborator on its own
config repo and has read on `internal/docs`, and nothing else. Labelling and
assigning an issue or a pull request in a project repo takes write access to
that repo, so the triage agent can't do either until the operator adds it as a

View file

@ -19,7 +19,7 @@ No MCP tool exists for config changes: opening the PR IS the request.
own config as well as how it changes it.
Branch protection (the agent isn't on the `main` push allowlist;
merge and approvals allowlist = `operators` team; see "Forge config repos"
below) makes the agent a write collaborator that **can't merge its own
below) makes the agent a read collaborator that **can't merge its own
config PR**.
2. An operator reviews the PR **on the forge** (native diff, threaded
comments, CI status) and merges it there. Merging needs membership of
@ -194,10 +194,12 @@ Every agent created through the swarm has a repo in the private
hive-c0re pushes nothing to it, keeps no clone and reads nothing from it: forge
`main` advances only by an operator's merge.
Each agent is a **write collaborator on its own** `agent-configs/<name>`
repo — so it can push a branch and open a config PR — but not a member
of any other agent's, so it can't reach another agent's config through
the forge. Branch protection keeps the agent off the `main` push
Each agent is a **read collaborator on its own** `agent-configs/<name>`
repo, so it can clone it and open a config PR through the AGit flow (no push
access needed), but not a member of any other agent's, so it can't reach
another agent's config through the forge. A grant of more than read, made by
an operator or before agents got read, is left as it is: the `AddRepoMember`
node only adds read where the agent has none. Branch protection keeps the agent off the `main` push
allowlist and allowlists merging to the `operators` team only, so an
agent can't fast-forward its own config or self-merge its PR (see
[Config changes](#config-changes) above). hive-c0re passes the credential to

View file

@ -63,8 +63,9 @@ live; tokens of other names are never touched.
Two things live in the `agent-configs` Forgejo organization:
- A config repo per agent (`agent-configs/<name>`). The
agent is a **write collaborator on its own** repo — it can push
config-change branches and open config PRs — but `main` is
agent is a **read collaborator on its own** repo: it clones it and
opens config PRs through the AGit flow (`hive-forge pr create --agit`),
which needs no push access. `main` is
branch-protected by swarm-controller: the merge and approval
allowlists are the `operators` team, and the agent can neither push
`main` directly nor self-merge. An operator's merge in the Forgejo UI

View file

@ -524,17 +524,6 @@ impl Client {
}
}
/// Add `user` as a collaborator on `CONFIG_ORG/repo` at `permission`.
async fn add_collaborator(
&self,
repo: &str,
user: &str,
permission: AddCollaboratorOptionPermission,
) -> Result<()> {
self.add_collaborator_to(CONFIG_ORG, repo, user, permission)
.await
}
/// Add `user` as a collaborator on `owner/repo` at `permission`.
/// Idempotent: re-adding an existing collaborator just updates its
/// permission (forgejo answers 204 either way; a 201 from older
@ -720,12 +709,14 @@ impl Client {
Ok(true)
}
/// Add `agent` as a **write** collaborator on `repo` (can push + open
/// PRs, cannot bypass branch protection) — the whole job of the
/// `AddRepoMember` node. Idempotent.
/// Add `agent` as a **read** collaborator on `repo` — the whole job of the
/// `AddRepoMember` node. Read is enough: the agent clones its config and
/// proposes a change through an `AGit` pull request, which needs no push
/// access. Idempotent, and it leaves a grant of read or more as it is, so a
/// write grant an operator made, or one made before this node granted
/// read, is never lowered.
pub async fn add_repo_member(&self, repo: &str, agent: &str) -> Result<()> {
self.add_collaborator(repo, agent, AddCollaboratorOptionPermission::Write)
.await
self.grant_read(CONFIG_ORG, repo, agent).await
}
/// Ensure `agent` exists as a Forgejo user account that cannot create

View file

@ -54,15 +54,21 @@ fn holds_read(p: &RepoCollaboratorPermission) -> bool {
impl Client {
/// Make `agent` a **read** collaborator on `internal/docs` — the whole job
/// of the `GrantDocsRead` node. Idempotent, and never lowers: an agent that
/// already holds read or more (write, admin, owner) is left as it is, since
/// the add endpoint would overwrite the permission. Needs the forge user to
/// exist.
/// of the `GrantDocsRead` node. Idempotent, and never lowers: see
/// [`Self::grant_read`]. Needs the forge user to exist.
pub async fn grant_docs_read(&self, agent: &str) -> Result<()> {
self.grant_read(SHARED_ORG, SHARED_DOCS_REPO, agent).await
}
/// Make `user` a **read** collaborator on `owner/repo` unless it already
/// holds read or more (write, admin, owner), which is left as it is: the
/// add endpoint would overwrite the permission, so adding at read would
/// lower a write grant an operator made.
pub(super) async fn grant_read(&self, owner: &str, repo: &str, user: &str) -> Result<()> {
match self
.api
.call(|api| {
api.repo_get_repo_permissions(SHARED_ORG, SHARED_DOCS_REPO, agent)
api.repo_get_repo_permissions(owner, repo, user)
.into_future()
})
.await
@ -71,18 +77,12 @@ impl Client {
Ok(_) => {}
Err(e) if is_not_found(&e) => {}
Err(e) => {
return Err(e).with_context(|| {
format!("read {agent}'s permission on {SHARED_ORG}/{SHARED_DOCS_REPO}")
});
return Err(e)
.with_context(|| format!("read {user}'s permission on {owner}/{repo}"));
}
}
self.add_collaborator_to(
SHARED_ORG,
SHARED_DOCS_REPO,
agent,
AddCollaboratorOptionPermission::Read,
)
.await
self.add_collaborator_to(owner, repo, user, AddCollaboratorOptionPermission::Read)
.await
}
/// Whether `agent` is a collaborator on `internal/docs` now.

View file

@ -40,7 +40,7 @@ pub(crate) enum SwarmNodeKind {
/// account this node hasn't created yet. See
/// `forge::Client::ensure_agent_user`.
CreateForgeUser { agent: String },
/// Add `agent` as a write collaborator on its own repo. See
/// Add `agent` as a read collaborator on its own repo. See
/// `forge::Client::add_repo_member`.
AddRepoMember { agent: String },
/// Add `agent` as a read collaborator on `internal/docs`. See