hyperhive/hive-c0re/src/forge/repos.rs

//! Repo + org plumbing on the local Forgejo: org / repo creation,
//! the meta + shared-docs + knowledge repos, per-agent config-repo
//! mirroring (`push_config` / `push_meta`), collaborator grants,
//! pull-mirrors, and branch-protection rules. The typed API-client
//! constructor + org-name constants live in the module root (`super`).

use std::path::Path;
use std::sync::Mutex;

use anyhow::{Context, Result};
use forgejo_api::structs::{
    AddCollaboratorOption, AddCollaboratorOptionPermission, CreateBranchProtectionOption,
    CreateOrgOption, CreateRepoOption, CreateTeamOption, CreateTeamOptionPermission,
    EditBranchProtectionOption, EditRepoOption, EditTeamOption, EditTeamOptionPermission,
    MigrateRepoOptions, MigrateRepoOptionsService, Repository,
};
use forgejo_api::{ApiErrorKind, ForgejoError};
use reqwest::StatusCode;

use crate::coordinator::Coordinator;

use super::{
    AGENTS_ORG, CONFIG_ORG, KNOWLEDGE_REPO, OPERATORS_TEAM, SHARED_DOCS_REPO, SHARED_ORG, api,
    core_auth_header, core_token, forge_git_url, forge_http_base, is_present,
};

/// Creation options for an empty repo defaulting to `main`.
fn repo_option(name: &str, private: bool) -> CreateRepoOption {
    CreateRepoOption {
        auto_init: Some(false),
        default_branch: Some("main".to_owned()),
        description: None,
        gitignores: None,
        issue_labels: None,
        license: None,
        name: name.to_owned(),
        object_format_name: None,
        private: Some(private),
        readme: None,
        template: None,
        trust_model: None,
    }
}

/// `EditRepoOption` with every field unset — repo edits only ever
/// change the one field the caller sets on top (Forgejo leaves `None`
/// fields untouched).
fn sparse_edit_repo_option() -> EditRepoOption {
    EditRepoOption {
        allow_fast_forward_only_merge: None,
        allow_manual_merge: None,
        allow_merge_commits: None,
        allow_rebase: None,
        allow_rebase_explicit: None,
        allow_rebase_update: None,
        allow_squash_merge: None,
        archived: None,
        autodetect_manual_merge: None,
        default_allow_maintainer_edit: None,
        default_branch: None,
        default_delete_branch_after_merge: None,
        default_merge_style: None,
        default_update_style: None,
        description: None,
        enable_prune: None,
        external_tracker: None,
        external_wiki: None,
        globally_editable_wiki: None,
        has_actions: None,
        has_issues: None,
        has_packages: None,
        has_projects: None,
        has_pull_requests: None,
        has_releases: None,
        has_wiki: None,
        ignore_whitespace_conflicts: None,
        internal_tracker: None,
        mirror_interval: None,
        name: None,
        private: None,
        template: None,
        website: None,
        wiki_branch: None,
    }
}

/// Whether a create-style call failed because the object already
/// exists. Forgejo signals this as HTTP 409 (conflict) or 422
/// (validation). The typed client surfaces those as
/// `ApiErrorKind::Other(409)` / `ValidationFailed` when the endpoint
/// spec lists the status, or as a bare `UnexpectedStatusCode`
/// otherwise — match all shapes defensively.
fn is_already_exists(e: &ForgejoError) -> bool {
    match e {
        ForgejoError::ApiError(api) => match api.error_kind() {
            ApiErrorKind::ValidationFailed => true,
            ApiErrorKind::Other(s) => *s == StatusCode::CONFLICT,
            _ => false,
        },
        ForgejoError::UnexpectedStatusCode(s) => {
            *s == StatusCode::CONFLICT || *s == StatusCode::UNPROCESSABLE_ENTITY
        }
        _ => false,
    }
}

/// Whether an error is specifically an HTTP 409 conflict (and NOT a
/// 422): the migrate endpoint's 422 is a validation error (bad
/// `clone_addr` / service) and must surface, so it can't share
/// [`is_already_exists`]'s 422 tolerance.
fn is_conflict(e: &ForgejoError) -> bool {
    match e {
        ForgejoError::ApiError(api) => {
            matches!(api.error_kind(), ApiErrorKind::Other(s) if *s == StatusCode::CONFLICT)
        }
        ForgejoError::UnexpectedStatusCode(s) => *s == StatusCode::CONFLICT,
        _ => false,
    }
}

/// Whether a rendered error message names the "already exists" case. The
/// discriminator that tells a *benign* already-exists 422 apart from a
/// *real* validation 422 (invalid units, etc.). Case-insensitive.
fn message_says_already_exists(rendered: &str) -> bool {
    rendered.to_lowercase().contains("already exists")
}

/// Whether `e` is Forgejo saying the resource already exists — matching
/// the 409 conflict shape *and* the 422 shape this Forgejo build actually
/// returns for a duplicate team: `validation failed: team already exists`.
/// A 409-only [`is_conflict`] check misses that 422, so the caller fires
/// a spurious "provisioning failed" warning every boot and skips its
/// settings reconcile. This still surfaces *other* 422s (bad request
/// body) as real failures — only a 422 whose message names the
/// already-exists case is folded in.
fn is_already_exists_lenient(e: &ForgejoError) -> bool {
    if is_conflict(e) {
        return true;
    }
    let is_unprocessable = match e {
        ForgejoError::ApiError(api) => matches!(api.error_kind(), ApiErrorKind::ValidationFailed),
        ForgejoError::UnexpectedStatusCode(s) => *s == StatusCode::UNPROCESSABLE_ENTITY,
        _ => false,
    };
    is_unprocessable && message_says_already_exists(&e.to_string())
}

/// Whether an error is Forgejo saying 404 — the resource is absent,
/// as opposed to a transport / auth / server failure.
fn is_not_found(e: &ForgejoError) -> bool {
    match e {
        ForgejoError::ApiError(api) => {
            matches!(api.error_kind(), ApiErrorKind::NotFound { .. })
        }
        ForgejoError::UnexpectedStatusCode(s) => *s == StatusCode::NOT_FOUND,
        _ => false,
    }
}

/// Fold a repo-creation result's "already exists" (409 / 422) into
/// success. `label` is `<owner>/<name>` — purely for log + error
/// context.
fn created_or_exists(res: Result<Repository, ForgejoError>, label: &str) -> Result<()> {
    match res {
        Ok(_) => {
            tracing::info!(%label, "forge: created repo");
            Ok(())
        }
        Err(e) if is_already_exists(&e) => {
            tracing::debug!(%label, "forge: repo already exists");
            Ok(())
        }
        Err(e) => Err(e).with_context(|| format!("create repo {label}")),
    }
}

/// Set an existing repo to public visibility. No-op if the repo is
/// already public. Used for `internal/knowledge` which may have been
/// created as private on an older deployment.
async fn set_repo_public(owner: &str, repo: &str, token: &str) -> Result<()> {
    let mut edit = sparse_edit_repo_option();
    edit.private = Some(false);
    api(token)?
        .repo_edit(owner, repo, edit)
        .await
        .with_context(|| format!("edit {owner}/{repo} (set public)"))?;
    tracing::debug!(%owner, %repo, "forge: repo set to public");
    Ok(())
}

/// Create `name` inside org `org` as a public repo. Idempotent.
async fn ensure_org_repo_public(org: &str, name: &str, token: &str) -> Result<()> {
    let res = api(token)?
        .create_org_repo(org, repo_option(name, false))
        .await;
    created_or_exists(res, &format!("{org}/{name}"))
}

/// Create a repo in the token-owner's own namespace. `token` belongs
/// to the user we want the repo owned by (we use `core`'s token for
/// `core/meta`). Idempotent.
pub async fn ensure_repo(name: &str, token: &str) -> Result<()> {
    let res = api(token)?
        .create_current_user_repo(repo_option(name, true))
        .await;
    created_or_exists(res, &format!("core/{name}"))
}

/// Create `name` inside org `org` (used for `agent-configs/<agent>`).
/// Idempotent.
async fn ensure_org_repo(org: &str, name: &str, token: &str) -> Result<()> {
    let res = api(token)?
        .create_org_repo(org, repo_option(name, true))
        .await;
    created_or_exists(res, &format!("{org}/{name}"))
}

/// Dashboard-warning guard for the meta non-fast-forward condition.
/// Held while `core/meta` remote is ahead of our local mirror; cleared
/// automatically when the next push succeeds.
static META_NON_FF_GUARD: Mutex<Option<crate::warnings::WarningGuard>> = Mutex::new(None);

/// Push `dir` (the meta repo) to `core/meta` on the local forge.
/// Best-effort: returns Err which callers log + ignore. No-op when
/// the core token isn't present yet (forge container not provisioned).
pub async fn push_meta(dir: &Path) -> Result<()> {
    let Some(token) = core_token() else {
        return Ok(());
    };
    // Token-in-URL push. Forgejo accepts `oauth2:<token>` or just
    // any-username:<token>; using `core` matches the owner so the
    // remote name is self-describing.
    //
    // No --force: `meta.rs` uses regular `git commit` (append-only),
    // so the push is always fast-forward in normal operation.  Per
    // operator directive, hive-c0re must not force-push anywhere.  If
    // the remote is somehow ahead (e.g. split-brain or manual push) we
    // raise a dashboard warning and leave the remote intact rather than
    // silently erasing its history — consistent with `push_config`'s
    // non-fast-forward handling.
    let url = forge_git_url("core/meta");
    let out = crate::lifecycle::git_command_authed(&core_auth_header(&token))
        .current_dir(dir)
        .args(["push", &url, "HEAD:main"])
        .output()
        .await
        .context("invoke git push core/meta")?;
    if !out.status.success() {
        let stderr = String::from_utf8_lossy(&out.stderr);
        if stderr.contains("non-fast-forward") || stderr.contains("fetch first") {
            let msg = "forge/core/meta push rejected (non-fast-forward): \
                       remote is ahead of local; leaving remote history intact";
            tracing::warn!("{msg}");
            // Raise a persistent dashboard banner; cleared on next successful push.
            if let Ok(mut g) = META_NON_FF_GUARD.lock() {
                *g = Some(crate::warnings::set_warning(
                    "forge-meta-non-ff",
                    "warn",
                    msg,
                ));
            }
            return Ok(());
        }
        anyhow::bail!(
            "git push core/meta failed ({}): {}",
            out.status,
            stderr.trim()
        );
    }
    // Successful push: clear any outstanding non-ff warning.
    if let Ok(mut g) = META_NON_FF_GUARD.lock() {
        *g = None;
    }
    tracing::info!("forge: pushed meta to core/meta");
    Ok(())
}

/// Ensure the `agent-configs/<name>` repo exists so the first
/// `push_config` doesn't 404, and wire it as the agent-editable PR surface:
/// the agent is a **write** collaborator (can push feature branches +
/// open config PRs) and `main` is branch-protected core-only (only hive-c0re's
/// merge handler lands on it; operator approval required). No-op when the forge
/// isn't running or the core token isn't minted yet. Safe to call on every
/// spawn and on every startup (all steps idempotent).
pub async fn ensure_config_repo(name: &str) -> Result<()> {
    if !is_present().await {
        return Ok(());
    }
    let Some(token) = core_token() else {
        return Ok(());
    };
    ensure_org_repo(CONFIG_ORG, name, &token).await?;
    // Agent = write collaborator: it can push config-PR branches + open PRs,
    // but the branch protection below keeps it off `main` directly.
    add_collaborator(
        CONFIG_ORG,
        name,
        name,
        AddCollaboratorOptionPermission::Write,
        &token,
    )
    .await?;
    // Protect `main` core-only, fast-forward-only (no auto force-push). A
    // failure here is security-relevant (an unprotected config repo lets an
    // operator-merged config PR bypass the deploy pipeline), so it's tracked
    // on the dashboard banner in addition to the journal warning callers
    // already log — see `record_branch_protection_result`.
    let result = apply_config_repo_branch_protection(name, &token).await;
    record_branch_protection_result(name, result.is_ok());
    result
}

/// Dashboard-banner tracker for [`apply_config_repo_branch_protection`]
/// failures. The registry ([`crate::warnings::set_warning`]) only takes
/// `&'static str` kinds, so a dynamic per-agent key isn't possible — instead
/// this keeps one static `crit` warning (`"branch_protection_missing"`) whose
/// message lists every agent currently failing to protect, and clears it
/// once the set is empty. Called on every `ensure_config_repo` pass (startup
/// sweep + per-rebuild), so a fixed agent drops out of the message on its
/// next successful sweep without requiring a restart.
fn record_branch_protection_result(name: &str, ok: bool) {
    use std::collections::BTreeSet;
    use std::sync::{OnceLock, PoisonError};

    use crate::warnings::{WarningGuard, set_warning};

    static FAILING: OnceLock<Mutex<BTreeSet<String>>> = OnceLock::new();
    static GUARD: OnceLock<Mutex<Option<WarningGuard>>> = OnceLock::new();

    let mut failing = FAILING
        .get_or_init(|| Mutex::new(BTreeSet::new()))
        .lock()
        .unwrap_or_else(PoisonError::into_inner);
    if ok {
        failing.remove(name);
    } else {
        failing.insert(name.to_owned());
    }

    let mut guard = GUARD
        .get_or_init(|| Mutex::new(None))
        .lock()
        .unwrap_or_else(PoisonError::into_inner);
    if failing.is_empty() {
        *guard = None;
        return;
    }
    let names = failing.iter().cloned().collect::<Vec<_>>().join(", ");
    let message = format!(
        "config-repo branch protection not applied for: {names} — \
         operator-merged config PRs for these agents could bypass the deploy pipeline"
    );
    match guard.as_ref() {
        Some(g) => g.update("crit", message),
        None => *guard = Some(set_warning("branch_protection_missing", "crit", message)),
    }
}

/// Ensure the `internal/docs` repo exists. Called once at startup
/// after `ensure_org(SHARED_ORG)`. Idempotent — `ensure_org_repo`
/// treats 409 as success.
pub async fn ensure_shared_docs_repo(core_token: &str) -> Result<()> {
    ensure_org_repo(SHARED_ORG, SHARED_DOCS_REPO, core_token).await
}

/// Grant agent `name` read-only collaborator access to `internal/docs`.
/// Idempotent: re-adding an existing collaborator succeeds (Forgejo
/// answers 204 either way). Mirrors `meta_read_access` so agents can
/// clone the shared docs repo without authentication hassle.
pub async fn shared_docs_access(name: &str, core_token: &str) -> Result<()> {
    add_collaborator(
        SHARED_ORG,
        SHARED_DOCS_REPO,
        name,
        AddCollaboratorOptionPermission::Read,
        core_token,
    )
    .await?;
    tracing::info!(%name, "forge: granted shared-docs read access");
    Ok(())
}

/// Ensure the `internal/knowledge` repo exists and is public.
/// Called once at startup after `ensure_org(SHARED_ORG)`. Idempotent.
///
/// The repo is created as public so any agent with a forge account can
/// fork it and open PRs to contribute. Existing deployments that ended
/// up with a private repo are patched to public on the next hive-c0re
/// startup via `set_repo_public`.
pub async fn ensure_knowledge_repo(core_token: &str) -> Result<()> {
    ensure_org_repo_public(SHARED_ORG, KNOWLEDGE_REPO, core_token).await?;
    // Ensure public even if the repo already existed as private (older deployment).
    set_repo_public(SHARED_ORG, KNOWLEDGE_REPO, core_token).await
}

/// Grant agent `name` read-only collaborator access to `core/meta` on
/// the forge so the agent can clone/fetch the meta flake. Idempotent:
/// re-adding an existing collaborator succeeds (Forgejo answers 204
/// either way).
pub async fn meta_read_access(name: &str, core_token: &str) -> Result<()> {
    add_collaborator(
        "core",
        "meta",
        name,
        AddCollaboratorOptionPermission::Read,
        core_token,
    )
    .await?;
    tracing::info!(%name, "forge: granted meta read access");
    Ok(())
}

/// Add the forge `core/meta.git` URL as the `meta` remote in the
/// agent's proposed config repo so the agent (and the manager) can
/// fetch the meta flake from the forge. Idempotent: no-op when the
/// remote already points at the right URL, or when the proposed repo
/// does not exist yet. No-op when the forge is not running.
pub async fn ensure_meta_remote(name: &str) -> Result<()> {
    if !is_present().await {
        return Ok(());
    }
    // A malformed name has no proposed config repo (repos are only created
    // under a validated Ident), so there's nothing to wire — no-op.
    let Ok(agent) = hive_types::Ident::parse(name) else {
        return Ok(());
    };
    let proposed_dir = Coordinator::agent_proposed_dir(&agent);
    if !proposed_dir.join(".git").exists() {
        return Ok(());
    }
    let want = format!("{}/core/meta.git", forge_http_base());
    let existing = crate::lifecycle::git_command()
        .current_dir(&proposed_dir)
        .args(["remote", "get-url", "meta"])
        .output()
        .await
        .context("git remote get-url meta")?;
    if existing.status.success() {
        let current = String::from_utf8_lossy(&existing.stdout).trim().to_owned();
        if current == want {
            return Ok(());
        }
        crate::lifecycle::git(&proposed_dir, &["remote", "set-url", "meta", &want]).await
    } else {
        crate::lifecycle::git(&proposed_dir, &["remote", "add", "meta", &want]).await
    }
}

/// Mirror agent `name`'s applied config repo — every status tag
/// (`proposal` / `approved` / `building` / `deployed` / `failed` /
/// `denied`) plus a best-effort `main` — to `agent-configs/<name>` on
/// the local forge. Best-effort: returns Err which callers log + ignore.
/// No-op when the forge isn't seeded or the applied repo doesn't exist.
///
/// Call this after every hive-c0re mutation of an applied repo's refs
/// so the forge copy always reflects what core actually did.
///
/// **Two separate pushes, not one.** The status tags are id-suffixed
/// (`deployed/<id>`, …) and add-only, so they must always land. `main`,
/// by contrast, is core-only branch-protected and authoritatively
/// advanced by `pr_merge`'s ff-only merge API — so a mirror push of an
/// already-established `main` is routinely rejected (protected-branch,
/// or non-fast-forward after a rolled-back deploy). Git's pre-receive
/// hook is all-or-nothing: bundling both refspecs in one push means that
/// `main` reject declines the whole push, dropping the tags too.
/// So we push the tags on their own first, then attempt `main`
/// separately and swallow the expected reject. The `main` push still
/// matters for the initial seed of a fresh (empty) config repo, where it
/// creates `main`. Operator-driven divergence fix: the `hivectl forge
/// reconcile-config` command.
///
/// The tokenised URL is passed straight to `git push` and deliberately
/// never stored as a named remote: the applied repo is bind-mounted
/// READ-ONLY into the manager container (`/applied`), so a token in
/// `.git/config` would leak core's admin credential to an agent.
pub async fn push_config(name: &str) -> Result<()> {
    let Some(token) = core_token() else {
        return Ok(());
    };
    let dir = crate::paths::applied_dir(name);
    if !dir.join(".git").exists() {
        return Ok(());
    }
    let url = forge_git_url(&format!("{CONFIG_ORG}/{name}"));
    let auth = core_auth_header(&token);
    // Tags first, in their own push, so they land regardless of main's fate.
    let out = run_config_push(&dir, &url, &auth, "refs/tags/*:refs/tags/*").await?;
    if !out.status.success() {
        anyhow::bail!(
            "git push tags {CONFIG_ORG}/{name} failed ({}): {}",
            out.status,
            String::from_utf8_lossy(&out.stderr).trim()
        );
    }
    // Then main on its own — a protected-branch / non-ff reject of an
    // established main is expected (pr_merge owns main); only the initial
    // empty-repo seed actually advances it here.
    let out = run_config_push(&dir, &url, &auth, "refs/heads/main:refs/heads/main").await?;
    if !out.status.success() {
        let stderr = String::from_utf8_lossy(&out.stderr);
        if stderr.contains("non-fast-forward")
            || stderr.contains("protected branch")
            || stderr.contains("pre-receive hook declined")
        {
            tracing::info!(
                %name,
                "forge: mirror push of main skipped — forge main is protected / \
                 owned by the config-PR ff-merge (expected); tags mirrored"
            );
            return Ok(());
        }
        anyhow::bail!(
            "git push main {CONFIG_ORG}/{name} failed ({}): {}",
            out.status,
            stderr.trim()
        );
    }
    tracing::info!(%name, "forge: mirrored applied config to agent-configs");
    Ok(())
}

/// Reseed a revived agent's `applied` repo — missing its `.git` entirely,
/// e.g. after `destroy --purge` — by fetching `agent-configs/<name>`'s
/// `main` from the forge, [`push_config`]'s mirror target. Unlike
/// `proposed` (seeded once at first spawn and never touched again — see
/// `lifecycle::setup_proposed`'s own doc comment), the forge mirror is
/// kept current after every deploy, so this reflects the agent's actual
/// last-deployed state rather than a stale creation-time snapshot.
///
/// Best-effort, mirroring [`push_config`]'s own shape: returns `false`
/// (not an error) rather than bailing when the forge is absent, the core
/// token isn't minted yet, `applied` already has a `.git` (nothing to
/// do), or any git step fails — callers fall back to a less-current
/// source or the original hard error. Only `true` once `applied` is
/// actually seeded and tagged `deployed/0`, mirroring first-spawn's own
/// `setup_applied` seed shape exactly.
pub async fn fetch_config_main_into_applied(name: &str) -> bool {
    if !is_present().await {
        return false;
    }
    let Some(token) = core_token() else {
        return false;
    };
    let applied = crate::paths::applied_dir(name);
    if applied.join(".git").exists() {
        return false;
    }
    if let Err(e) = std::fs::create_dir_all(&applied) {
        tracing::warn!(%name, error = ?e, "forge: applied reseed mkdir failed");
        return false;
    }
    if let Err(e) = crate::lifecycle::git(&applied, &["init", "--initial-branch=main"]).await {
        tracing::warn!(%name, error = ?e, "forge: applied reseed init failed");
        return false;
    }
    let url = forge_git_url(&format!("{CONFIG_ORG}/{name}"));
    let auth = core_auth_header(&token);
    let out = crate::lifecycle::git_command_authed(&auth)
        .current_dir(&applied)
        .args([
            "fetch",
            "--no-tags",
            "--update-head-ok",
            &url,
            "refs/heads/main:refs/heads/main",
        ])
        .output()
        .await;
    match out {
        Ok(o) if o.status.success() => {}
        Ok(o) => {
            tracing::warn!(
                %name,
                stderr = %String::from_utf8_lossy(&o.stderr).trim(),
                "forge: applied reseed fetch failed"
            );
            return false;
        }
        Err(e) => {
            tracing::warn!(%name, error = ?e, "forge: applied reseed fetch failed");
            return false;
        }
    }
    if let Err(e) = crate::lifecycle::git_read_tree_reset(&applied, "refs/heads/main").await {
        tracing::warn!(%name, error = ?e, "forge: applied reseed read-tree failed");
        return false;
    }
    if let Err(e) = crate::lifecycle::git_tag(&applied, "deployed/0", "refs/heads/main").await {
        tracing::warn!(%name, error = ?e, "forge: applied reseed tag failed");
        return false;
    }
    tracing::info!(%name, "forge: reseeded applied repo from agent-configs main");
    true
}

/// Seed an agent's `proposed` repo from `agent-configs/<name>` on the forge.
///
/// When the swarm creates an agent it writes that agent's config to the forge
/// before any hive is told to deploy it, so the hive's job here is to **take**
/// that config rather than author a second one beside it. The local template
/// is the fallback, not the default.
///
/// Best-effort like [`fetch_config_main_into_applied`], and `false` is an
/// ordinary answer rather than an error: the forge is absent, the core token
/// is not minted yet, or `agent-configs/<name>` has no `main` — the last of
/// which is the *normal* case for the hive-level create flow, where the repo
/// is only created after the first spawn (`actions::forge_after_first_spawn`).
///
/// `clone` rather than `init` + `fetch`: git removes a directory it created
/// when the clone fails, and cloning into an existing empty dir leaves no
/// `.git` behind either — so the caller's "does this repo already exist"
/// check reads exactly the same before and after a failed attempt.
pub async fn clone_config_into_proposed(dir: &Path, name: &str) -> bool {
    if !is_present().await {
        return false;
    }
    let Some(token) = core_token() else {
        return false;
    };
    let url = forge_git_url(&format!("{CONFIG_ORG}/{name}"));
    // `--branch main` fails outright against a repo with no commits, so an
    // empty repo reads as "nothing to take" instead of cloning to an unborn
    // HEAD that would then look like a seeded checkout.
    let out = crate::lifecycle::git_command_authed(&core_auth_header(&token))
        .args([
            "clone",
            "--branch",
            "main",
            &url,
            &dir.display().to_string(),
        ])
        .output()
        .await;
    match out {
        Ok(o) if o.status.success() => {
            tracing::info!(%name, "forge: seeded proposed repo from agent-configs main");
            true
        }
        Ok(o) => {
            // `info`, not `warn`: on the hive-level create flow every first
            // spawn lands here and nothing is wrong. Logged at all because
            // "where did this agent's config come from" is worth answering
            // from the journal, and proposed is seeded exactly once.
            tracing::info!(
                %name,
                stderr = %String::from_utf8_lossy(&o.stderr).trim(),
                "forge: no config to clone for this agent; seeding the template"
            );
            false
        }
        Err(e) => {
            tracing::warn!(%name, error = ?e, "forge: proposed clone failed to run");
            false
        }
    }
}

/// Run a single `git push <url> <refspec>` in the applied repo `dir` and
/// return the raw output for the caller to classify. Split out so
/// [`push_config`] can push tags and `main` as independent pushes.
async fn run_config_push(
    dir: &std::path::Path,
    url: &str,
    auth: &str,
    refspec: &str,
) -> Result<std::process::Output> {
    crate::lifecycle::git_command_authed(auth)
        .current_dir(dir)
        .args(["push", url, refspec])
        .output()
        .await
        .context("invoke git push agent-configs")
}

/// Create an org named `name` (`org_create`). Idempotent: HTTP 422
/// ("user already exists") / 409 is treated as success.
pub(super) async fn ensure_org(name: &str, admin_token: &str) -> Result<()> {
    let org = CreateOrgOption {
        description: None,
        email: None,
        full_name: None,
        location: None,
        repo_admin_change_team_access: None,
        username: name.to_owned(),
        visibility: None,
        website: None,
    };
    match api(admin_token)?.org_create(org).await {
        Ok(_) => {
            tracing::info!(%name, "forge: created org");
            Ok(())
        }
        Err(e) if is_already_exists(&e) => {
            tracing::debug!(%name, "forge: org already exists");
            Ok(())
        }
        Err(e) => Err(e).with_context(|| format!("create org {name}")),
    }
}

/// One operator-declared pull-mirror, forwarded from the nix
/// `services.hyperhive.forge.mirrors` option as JSON in
/// `HYPERHIVE_FORGE_MIRRORS`.
#[derive(serde::Deserialize)]
struct Mirror {
    /// Upstream clone URL to mirror from (e.g. `https://github.com/actions/checkout`).
    upstream: String,
    /// Local `<owner>/<repo>` the mirror is created at.
    dest: String,
}

/// Ensure each `HYPERHIVE_FORGE_MIRRORS` entry exists as a real Forgejo
/// pull-mirror. The env carries the JSON-encoded nix `forge.mirrors` list
/// (plus the CI-auto `actions/checkout` entry). Absent/empty env = no-op.
/// Per-mirror failures warn and continue — never abort the startup sweep.
pub(super) async fn ensure_mirrors(admin_token: &str) {
    let raw = match std::env::var("HYPERHIVE_FORGE_MIRRORS") {
        Ok(s) if !s.trim().is_empty() => s,
        _ => return,
    };
    let mirrors: Vec<Mirror> = match serde_json::from_str(&raw) {
        Ok(m) => m,
        Err(e) => {
            tracing::warn!(error = ?e, "forge: HYPERHIVE_FORGE_MIRRORS is not valid JSON; skipping mirror seed");
            return;
        }
    };
    for m in mirrors {
        let Some((owner, repo)) = m.dest.split_once('/') else {
            tracing::warn!(dest = %m.dest, "forge: mirror dest is not <owner>/<repo>; skipping");
            continue;
        };
        // Create the dest org first (idempotent); the mirror can't land
        // without its owner existing.
        if let Err(e) = ensure_org(owner, admin_token).await {
            tracing::warn!(%owner, error = ?e, "forge: ensure_org for mirror failed");
            continue;
        }
        if let Err(e) = ensure_mirror_repo(&m.upstream, owner, repo, admin_token).await {
            tracing::warn!(dest = %m.dest, error = ?e, "forge: ensure_mirror_repo failed");
        }
    }
}

/// Periodic sync interval for pull-mirrors. Forgejo syncs mirrors
/// on-access by default, which re-introduces external DNS latency on
/// every `git clone` (the hive-ci runner shares the host netns and is
/// therefore affected by host resolver blips). A fixed periodic interval
/// isolates CI from transient DNS failures — a stale mirror is
/// acceptable; a broken clone because of a momentary DNS blip is not.
const MIRROR_INTERVAL: &str = "8h0m0s";

/// Create `owner/repo` as a pull-mirror of `upstream` via the migrate API.
/// Idempotent: if the repo already exists this function patches its
/// `mirror_interval` to ensure it matches (covers mirrors that were
/// created before the interval was introduced). A 409 on the migrate
/// call (a race between the existence check and the migrate) is also
/// success.
async fn ensure_mirror_repo(
    upstream: &str,
    owner: &str,
    repo: &str,
    admin_token: &str,
) -> Result<()> {
    let client = api(admin_token)?;
    match client.repo_get(owner, repo).await {
        Ok(_) => {
            // Mirror already present. Patch interval so mirrors seeded before
            // this field was introduced (or with a different value) converge.
            let mut edit = sparse_edit_repo_option();
            edit.mirror_interval = Some(MIRROR_INTERVAL.to_owned());
            match client.repo_edit(owner, repo, edit).await {
                Ok(_) => {
                    tracing::debug!(%owner, %repo, interval = MIRROR_INTERVAL, "forge: pull-mirror interval updated");
                }
                Err(e) => {
                    tracing::warn!(
                        %owner, %repo, error = %e,
                        "forge: failed to set mirror_interval on existing pull-mirror"
                    );
                }
            }
            return Ok(());
        }
        // Absent — fall through to migrate.
        Err(e) if is_not_found(&e) => {}
        // Anything else (transport, auth, 5xx) leaves the repo's existence
        // unknown: migrating anyway would fold a 409 into success and skip
        // the interval patch this pass. Surface it instead.
        Err(e) => {
            return Err(e).with_context(|| format!("get pull-mirror {owner}/{repo}"));
        }
    }
    let opts = MigrateRepoOptions {
        auth_password: None,
        auth_token: None,
        auth_username: None,
        clone_addr: upstream.to_owned(),
        description: None,
        issues: None,
        labels: None,
        lfs: None,
        lfs_endpoint: None,
        milestones: None,
        mirror: Some(true),
        // Periodic refresh instead of on-access sync — keeps CI isolated
        // from external DNS failures at clone time.
        mirror_interval: Some(MIRROR_INTERVAL.to_owned()),
        private: Some(false),
        pull_requests: None,
        releases: None,
        repo_name: repo.to_owned(),
        repo_owner: Some(owner.to_owned()),
        service: Some(MigrateRepoOptionsService::Git),
        uid: None,
        wiki: None,
    };
    match client.repo_migrate(opts).await {
        Ok(_) => {
            tracing::info!(%owner, %repo, %upstream, interval = MIRROR_INTERVAL, "forge: created pull-mirror");
            Ok(())
        }
        // 409 = a race created it between our existence check and here (the
        // check is the real idempotency guard). NOT 422: for the migrate
        // endpoint 422 is a validation error (bad clone_addr / service), so
        // it must surface via the error arm, not be swallowed as "already
        // exists" — hence `is_conflict`, not `is_already_exists`.
        Err(e) if is_conflict(&e) => {
            tracing::debug!(%owner, %repo, "forge: pull-mirror already exists (race)");
            Ok(())
        }
        Err(e) => Err(e).with_context(|| format!("migrate pull-mirror {owner}/{repo}")),
    }
}

/// Provision the [`OPERATORS_TEAM`] inside `org` as an **empty** team.
/// Branch protection on that org's repos references it as the
/// merge/approval whitelist; the operator adds herself as a member via the
/// forge UI / hivectl. `includes_all_repositories` so the gate applies to
/// every repo in the org; `write` is enough to approve + merge. hive-c0re
/// never manages membership. Idempotent (409 = already exists).
///
/// Must run for BOTH [`AGENTS_ORG`] and [`CONFIG_ORG`]: Gitea teams are
/// org-scoped, so a config-repo branch-protection rule referencing
/// `operators` needs the team to exist in `agent-configs` too. Missing it
/// there 422'd every `apply_config_repo_branch_protection`, leaving config
/// repos unprotected — operator-merged config PRs then bypassed the deploy
/// pipeline and silently didn't apply.
///
/// Uses `is_conflict` (409 only) — NOT `is_already_exists` (which also
/// folds 422 into "already exists"). A 422 from `org_create_team` is a
/// real validation error (bad request shape, missing units, etc.) that
/// must surface so it can be fixed; the previous 422-swallowing hid the
/// true cause and left the team silently uncreated every boot.
/// Repo-unit access flags for the `operators` team.
/// Explicit list so Forgejo doesn't reject a null/absent `units` field;
/// a `write`-permission team needs at least `repo.code` + `repo.pulls`
/// to review and merge PRs.
const OPERATORS_TEAM_UNITS: &[&str] = &[
    "repo.code",
    "repo.issues",
    "repo.pulls",
    "repo.releases",
    "repo.wiki",
    "repo.projects",
    "repo.packages",
];

pub(super) async fn ensure_operators_team(org: &str, token: &str) -> Result<()> {
    let units_vec: Vec<String> = OPERATORS_TEAM_UNITS.iter().map(|&s| s.to_owned()).collect();
    let team = CreateTeamOption {
        can_create_org_repo: Some(false),
        description: Some("hyperhive operators — merge gate for agent repos".to_owned()),
        includes_all_repositories: Some(true),
        name: OPERATORS_TEAM.to_owned(),
        permission: Some(CreateTeamOptionPermission::Write),
        units: Some(units_vec.clone()),
        units_map: None,
    };
    let client = api(token)?;
    match client.org_create_team(org, team).await {
        Ok(_) => {
            tracing::info!(%org, "forge: created {OPERATORS_TEAM} team");
            Ok(())
        }
        // Team already exists — reconcile settings to desired state so a
        // team created with an older/wrong shape self-heals on next boot.
        // Forgejo signals the duplicate as a 409 conflict OR (this build) a
        // 422 `validation failed: team already exists`; both mean the same
        // thing, so fold both in via `is_already_exists_lenient` — a 409-only
        // check missed the 422 and warned every boot. List teams to find the
        // id (required by org_edit_team), then unconditionally PATCH to the
        // desired settings. Members are a separate endpoint; untouched here.
        Err(e) if is_already_exists_lenient(&e) => {
            let (_headers, teams) = client
                .org_list_teams(org)
                .await
                .with_context(|| format!("list teams for {org}"))?;
            let team_id = teams
                .into_iter()
                .find(|t| t.name.as_deref() == Some(OPERATORS_TEAM))
                .and_then(|t| t.id)
                .with_context(|| {
                    format!("{OPERATORS_TEAM} team not found in {org} after 409 conflict")
                })?;
            let edit = EditTeamOption {
                can_create_org_repo: Some(false),
                description: Some("hyperhive operators — merge gate for agent repos".to_owned()),
                includes_all_repositories: Some(true),
                name: OPERATORS_TEAM.to_owned(),
                permission: Some(EditTeamOptionPermission::Write),
                units: Some(units_vec),
                units_map: None,
            };
            client
                .org_edit_team(team_id, edit)
                .await
                .with_context(|| format!("reconcile {org}/{OPERATORS_TEAM} team settings"))?;
            tracing::debug!(%org, "forge: reconciled {OPERATORS_TEAM} team settings");
            Ok(())
        }
        // Any OTHER error (a 422 that is NOT already-exists, or a transport
        // / auth failure): surface it — don't mask a real failure. A
        // non-already-exists 422 means the request body is invalid (e.g.
        // Forgejo rejected the units list); it repeats every boot until fixed.
        Err(e) => Err(e).with_context(|| format!("create team {org}/{OPERATORS_TEAM}")),
    }
}

/// Add `user` as a collaborator on `owner/repo` at `permission`.
/// Idempotent: adding an existing collaborator just updates its
/// permission (Forgejo answers 204 either way; a 201 from older
/// versions is tolerated defensively).
async fn add_collaborator(
    owner: &str,
    repo: &str,
    user: &str,
    permission: AddCollaboratorOptionPermission,
    token: &str,
) -> Result<()> {
    let res = api(token)?
        .repo_add_collaborator(
            owner,
            repo,
            user,
            AddCollaboratorOption {
                permission: Some(permission),
            },
        )
        .await;
    match res {
        Ok(()) => {}
        Err(ForgejoError::UnexpectedStatusCode(s)) if s == StatusCode::CREATED => {}
        Err(e) => {
            return Err(e).with_context(|| format!("add collaborator {user} to {owner}/{repo}"));
        }
    }
    tracing::debug!(%owner, %repo, %user, ?permission, "forge: collaborator set");
    Ok(())
}

/// `CreateBranchProtectionOption` protecting `main` with every other
/// field unset — Forgejo treats `None` fields as their defaults, same
/// as the sparse JSON bodies the raw-HTTP predecessor sent. Callers
/// set the whitelist/approval fields they need on top.
fn main_branch_protection_option() -> CreateBranchProtectionOption {
    CreateBranchProtectionOption {
        apply_to_admins: None,
        approvals_whitelist_teams: None,
        approvals_whitelist_username: None,
        block_on_official_review_requests: None,
        block_on_outdated_branch: None,
        block_on_rejected_reviews: None,
        branch_name: Some("main".to_owned()),
        dismiss_stale_approvals: None,
        enable_approvals_whitelist: None,
        enable_merge_whitelist: None,
        enable_push: None,
        enable_push_whitelist: None,
        enable_status_check: None,
        ignore_stale_approvals: None,
        merge_whitelist_teams: None,
        merge_whitelist_usernames: None,
        protected_file_patterns: None,
        push_whitelist_deploy_keys: None,
        push_whitelist_teams: None,
        push_whitelist_usernames: None,
        require_signed_commits: None,
        required_approvals: None,
        rule_name: None,
        status_check_contexts: None,
        unprotected_file_patterns: None,
    }
}

/// Apply the operator merge-gate branch protection to `repo`'s default
/// branch: only [`OPERATORS_TEAM`] members can merge, and an
/// approving review from that team is required — so the author (a write-level
/// agent, not in the team) cannot merge its own PR.
///
/// Idempotent, but **verify-don't-trust**: a create failure is ambiguous —
/// "rule already exists" (success) OR a silent rejection that created NO rule
/// (e.g. a 422 where `OPERATORS_TEAM` doesn't exist in `AGENTS_ORG`). The old
/// code folded 200/409/422 into `Ok` and left the repo unprotected with no
/// error — a fail-open merge gate. So on any create error, GET the `main` rule
/// and only treat it as success if the rule is actually present (the exact fix
/// already applied to [`apply_config_repo_branch_protection`]).
async fn apply_operator_branch_protection(repo: &str, token: &str) -> Result<()> {
    let client = api(token)?;
    let mut rule = main_branch_protection_option();
    rule.enable_merge_whitelist = Some(true);
    rule.merge_whitelist_teams = Some(vec![OPERATORS_TEAM.to_owned()]);
    rule.enable_approvals_whitelist = Some(true);
    rule.approvals_whitelist_teams = Some(vec![OPERATORS_TEAM.to_owned()]);
    rule.required_approvals = Some(1);
    rule.block_on_official_review_requests = Some(true);
    let Err(create_err) = client
        .repo_create_branch_protection(AGENTS_ORG, repo, rule)
        .await
    else {
        tracing::info!(%repo, "forge: applied operator branch protection");
        return Ok(());
    };
    match client
        .repo_get_branch_protection(AGENTS_ORG, repo, "main")
        .await
    {
        Ok(_) => {
            tracing::debug!(
                %repo, create_error = %create_err,
                "forge: operator branch protection already present"
            );
            Ok(())
        }
        Err(check_err) => anyhow::bail!(
            "branch protection for {AGENTS_ORG}/{repo} not applied: create failed \
             ({create_err}); GET main rule failed ({check_err}), no `main` rule present"
        ),
    }
}

/// Apply branch protection to an `agent-configs/<name>` repo's `main` so it
/// can serve as the agent-editable, PR-merge config surface:
/// - **`main` is never directly pushable** — no push is enabled on the
///   protected branch, so neither the agent (a write collaborator) nor
///   hive-c0re can `git push` it. It only advances via the config-PR merge
///   node (`actions::run_deploy_apply`), which fast-forward-*merges* the reviewed
///   head through the forge merge API (`Do=fast-forward-only`,
///   `head_commit_id` pinned to the reviewed sha).
/// - **merge is whitelisted to `core`** — only hive-c0re can merge a config PR;
///   the agent can push feature branches + open PRs but can't land them.
/// - **the operator's dashboard approval is the gate** — approval happens on
///   the `MergeConfigPr` card and hive-c0re only merges an approved PR. There's
///   deliberately no Forgejo `required_approvals` review requirement: the flow
///   never does an in-forge review, so requiring one would only dead-block the
///   `core` merge. The dashboard approval + the `core`-only merge whitelist are
///   the real gate.
/// - **fast-forward-only** — `main` only ever advances by fast-forward; a raced
///   non-ff `main` is refused by the merge API rather than force-moved.
///
/// Idempotent: an existing rule for the branch (200/409/422) is success.
async fn apply_config_repo_branch_protection(repo: &str, token: &str) -> Result<()> {
    let client = api(token)?;
    let mut rule = main_branch_protection_option();
    // Only `core` may merge (through the config-PR merge API); no push is
    // enabled at all, so `main` can only advance via that merge. Push +
    // approval defaults are off in `main_branch_protection_option`, so a fresh
    // rule needs nothing but the merge whitelist. (`config_repo_protection_edit`
    // must clear the old push/approval fields explicitly, since a PATCH leaves
    // unset fields untouched.)
    rule.enable_merge_whitelist = Some(true);
    rule.merge_whitelist_usernames = Some(vec!["core".to_owned()]);
    let Err(create_err) = client
        .repo_create_branch_protection(CONFIG_ORG, repo, rule)
        .await
    else {
        tracing::info!(%repo, "forge: applied config-repo branch protection");
        return Ok(());
    };
    // Create failed. This is ambiguous — "rule already exists" (the common
    // idempotent case) OR a silent rejection that created no rule. Either way a
    // create never *updates* an existing rule, and repos protected under an
    // older shape (the push-based / approval-gated rules) carry stale settings.
    // So converge the existing rule with a PATCH that explicitly clears them:
    // it fixes those stale repos on the next boot's `ensure_config_repo` pass,
    // is a harmless no-op when the rule is already correct, and still fails
    // loudly when no rule can be established (you can't edit a rule that isn't
    // there) — so it can't leave a repo silently unprotected.
    match client
        .repo_edit_branch_protection(CONFIG_ORG, repo, "main", config_repo_protection_edit())
        .await
    {
        Ok(_) => {
            // Debug, not info: this PATCH runs on every `ensure_config_repo`
            // boot pass for every existing repo (create → 409 → converge), so
            // it's almost always a no-op re-assertion — info-logging it would
            // be N lines of noise per boot on a many-agent hive.
            tracing::debug!(
                %repo, create_error = %create_err,
                "forge: converged existing config-repo branch protection"
            );
            Ok(())
        }
        Err(edit_err) => anyhow::bail!(
            "branch protection for {CONFIG_ORG}/{repo} not applied: create failed \
             ({create_err}); edit of existing `main` rule failed ({edit_err})"
        ),
    }
}

/// The `EditBranchProtectionOption` that converges an existing
/// `agent-configs/<name>` `main` rule to the current desired shape: merge
/// whitelisted to `core`, **no direct push at all**, and **no in-forge approval
/// requirement** (the dashboard approval + `core`-only merge whitelist are the
/// gate). The push/approval fields are set to their explicit off-values, not
/// left `None`, so a repo carrying an older push-based or approval-gated rule is
/// actually *converged* rather than merely re-asserted — a PATCH leaves unset
/// fields untouched. Every field unrelated to this policy stays `None`.
fn config_repo_protection_edit() -> EditBranchProtectionOption {
    EditBranchProtectionOption {
        apply_to_admins: None,
        approvals_whitelist_teams: None,
        approvals_whitelist_username: None,
        block_on_official_review_requests: Some(false),
        block_on_outdated_branch: None,
        block_on_rejected_reviews: None,
        dismiss_stale_approvals: None,
        enable_approvals_whitelist: Some(false),
        enable_merge_whitelist: Some(true),
        enable_push: Some(false),
        enable_push_whitelist: Some(false),
        enable_status_check: None,
        ignore_stale_approvals: None,
        merge_whitelist_teams: None,
        merge_whitelist_usernames: Some(vec!["core".to_owned()]),
        protected_file_patterns: None,
        push_whitelist_deploy_keys: None,
        push_whitelist_teams: None,
        push_whitelist_usernames: Some(Vec::new()),
        require_signed_commits: None,
        required_approvals: Some(0),
        status_check_contexts: None,
        unprotected_file_patterns: None,
    }
}

/// Create a repo for `agent` in the c0re-owned [`AGENTS_ORG`] and wire the
/// perms: the org owns it (perms stay c0re-managed), the agent is added
/// as a **write** collaborator (not owner — can push + open PRs but can't
/// bypass branch protection), and the default branch gets the operator
/// merge gate. This is the sanctioned create path now that agents can't
/// create repos directly (`max_repo_creation = 0`). Idempotent.
pub async fn create_agent_repo(agent: &str, repo: &str, core_token: &str) -> Result<String> {
    ensure_org_repo(AGENTS_ORG, repo, core_token).await?;
    add_collaborator(
        AGENTS_ORG,
        repo,
        agent,
        AddCollaboratorOptionPermission::Write,
        core_token,
    )
    .await?;
    apply_operator_branch_protection(repo, core_token).await?;
    tracing::info!(%agent, %repo, "forge: created agent repo in {AGENTS_ORG} with operator merge gate");
    Ok(format!("{AGENTS_ORG}/{repo}"))
}

#[cfg(test)]
mod tests {
    use super::message_says_already_exists;

    #[test]
    fn already_exists_message_is_recognised() {
        // The exact 422 body this Forgejo build returns for a duplicate team.
        assert!(message_says_already_exists(
            "validation failed: team already exists [org_id: 10, name: operators]"
        ));
        // Case-insensitive.
        assert!(message_says_already_exists("Repository Already Exists"));
    }

    #[test]
    fn real_validation_error_is_not_treated_as_already_exists() {
        // A genuine bad-request 422 must still surface, not be folded in.
        assert!(!message_says_already_exists(
            "validation failed: units must not be empty"
        ));
        assert!(!message_says_already_exists("not found"));
    }
}