hyperhive/swarm-controller/src/forge.rs

//! Swarm-controller's own forgejo client. Self-contained — deliberately
//! not sharing code with `hive-c0re::forge` across the crate boundary
//! (forcing that split now, over a handful of idempotent CRUD-ish
//! calls, is premature plumbing; factor out a shared crate later if the
//! duplication actually starts to hurt).
//!
//! Reads its identity from the credentials `swarm-controller.nix`'s
//! `forgeEnv` wires in (`SWARM_CONTROLLER_FORGE_URL` +
//! `SWARM_CONTROLLER_FORGE_TOKEN_FILE`, the latter pointing at a
//! systemd `LoadCredential`-delivered file) — see that module and
//! `hive-forge/default.nix`'s `forgejo-swarm-controller-account` /
//! `hive-forge-swarm-controller-token` units for how the token itself
//! gets minted and delivered. Absent env (forge not configured on this
//! host) means [`Client::from_env`] returns `None` — every caller
//! treats that as "no forge access here" and continues, the same
//! graceful-absence shape the queue coordinates already use.

use anyhow::{Context, Result};
use forgejo_api::structs::{
    AddCollaboratorOption, AddCollaboratorOptionPermission, ChangeFileOperation,
    ChangeFileOperationOperation, ChangeFilesOptions, CreateBranchProtectionOption,
    CreateHookOption, CreateHookOptionConfig, CreateHookOptionType, CreateRepoOption,
    CreateUserOption, IssueListIssuesQuery, IssueListIssuesQueryState, IssueListIssuesQueryType,
    RepoGetContentsQuery, RepoListPullRequestsQuery, RepoListPullRequestsQueryState,
    RepoSearchQuery, StateType,
};
use forgejo_api::{ApiErrorKind, Auth, Forgejo, ForgejoError};
use futures_util::{StreamExt as _, TryStreamExt as _};
use reqwest::StatusCode;
use serde::{Deserialize, Serialize};
use std::collections::{BTreeMap, HashMap, HashSet};
use utoipa::ToSchema;

use crate::webhook::DeliveryKind;

/// An agent's open config-PR, as [`Client::list_open_config_prs`] reports it
/// and `GET /api/agents/{name}/config-pr` serves it.
#[derive(Clone, Debug, PartialEq, Serialize, Deserialize, ToSchema)]
pub struct ConfigPrStatus {
    pub pr_number: u64,
    /// Absent only if Forgejo itself omitted the field — every real PR has
    /// one; not worth failing the whole scan over.
    pub html_url: Option<String>,
}

/// One row of [`Client::issue_report`] — an open issue plus the two facts
/// derived from its dependency graph that swarm-ui's issue-report page
/// sorts/filters on. See that function's doc comment for how both are
/// computed.
#[derive(Clone, Debug, PartialEq, Serialize, ToSchema)]
pub struct IssueReportRow {
    /// `owner/name` — present on every row (not just the cross-repo
    /// aggregate) so a caller can use the same row type for either
    /// endpoint without a second shape to branch on.
    pub repo: String,
    pub number: i64,
    pub title: String,
    pub labels: Vec<String>,
    /// All assignees, not just one — forgejo's `Issue::assignee` is a
    /// legacy single-value field; `assignee`**s** (plural) is the real
    /// multi-assignee list, and this repo's issues do use more than one.
    pub assignees: Vec<String>,
    pub html_url: Option<String>,
    /// True if any of this issue's own dependencies is still open — same
    /// semantics `hive-forge issue dependency list` already uses (a closed
    /// blocker doesn't count).
    pub blocked: bool,
    /// How many OTHER issues in this same report list this issue as a
    /// dependency — the reverse of `blocked`. Ranks "fix this one to
    /// unblock the most other work" highest.
    pub depended_on_by_count: u32,
    /// How many distinct OTHER issues this issue unblocks, following the
    /// same reverse-dependency edges transitively (a depends on b depends
    /// on c: fixing c eventually frees both a and b, so c's count includes
    /// both). A superset of `depended_on_by_count`, which only counts the
    /// direct edge. See [`Client::issue_report`] for how it's walked.
    pub transitively_blocks_count: u32,
}

/// The `operators` team, whitelisted for the merge gate on every repo
/// this client protects — provisioned by `hive-c0re::forge::repos`
/// already (`ensure_operators_team`), not re-provisioned here. If that
/// assumption ever breaks (this becomes reachable before any hive's
/// `hive-c0re` has run its startup sweep), branch-protection creation
/// below will fail loudly rather than silently no-op — see
/// [`Client::create_repo`]'s doc comment.
const OPERATORS_TEAM: &str = "operators";

/// The org owning per-agent config repos — where this daemon creates them,
/// seeds them, and where the `pull_request` hook lives. Same value as
/// `hive-c0re::forge::CONFIG_ORG` — one forge, one org.
///
/// ⚠️ Duplicated across the crate boundary (this crate deliberately does not
/// depend on `hive-c0re`), so nothing makes the two fail together. The
/// failure mode if they drift is quiet: the hook is created on an org that
/// exists, forgejo reports it healthy, and it simply never fires.
///
/// ⚠️ **Not `agents`.** That org exists and is a different thing: it is the
/// namespace repos an *agent asks for* land in, per
/// `hive-c0re::forge::AGENTS_ORG`. Agent **config** repos have always lived
/// here, which is where `hive-c0re` reconciles, merges and mirrors them — a
/// config repo created in `agents` is invisible to every one of those paths,
/// and nothing errors, because both orgs exist and both accept a repo.
///
/// The merge gate survives the distinction: `hive-c0re` provisions the
/// `operators` team in **both** orgs precisely so branch protection can be
/// applied in either.
const CONFIG_ORG: &str = "agent-configs";

/// The hive-wide knowledge repo, where the `push` hook lives. Same values as
/// `hive-c0re::workers::knowledge::{ORG, REPO}`.
const KNOWLEDGE_ORG: &str = "internal";
const KNOWLEDGE_REPO: &str = "knowledge";

/// This daemon's own forge account — same constant as
/// `swarm-controller.nix`'s `swarmControllerForgeUser` (`hive-forge/default.nix`).
/// Needed in the branch-protection push whitelist (see
/// [`Client::apply_operator_branch_protection`]): being a site admin does
/// not exempt a caller from a protected branch's push check on the
/// content-edit API path (`repo_change_files`, what [`Client::seed_agent_config`]
/// uses) the way it does for a real `git push` — confirmed live against a
/// re-test where the account was already promoted to admin and the seed
/// commit still failed `ErrUserCannotCommit`. Whitelisting the account by
/// name is the same shape a git-push-based seed gets for free from the
/// admin bypass, without adding a `git` shell-out to this daemon.
const SWARM_CONTROLLER_FORGE_USER: &str = "swarm-controller";

/// The env vars `swarm-controller.nix`'s `forgeEnv` sets. Named here
/// once so [`Client::from_env`] and the nix module can't drift silently
/// — a rename on one side without the other fails loudly (env var
/// absent) rather than silently losing forge access.
const URL_ENV: &str = "SWARM_CONTROLLER_FORGE_URL";
const TOKEN_FILE_ENV: &str = "SWARM_CONTROLLER_FORGE_TOKEN_FILE";

/// Typed forgejo client, built once at startup from env.
pub struct Client {
    api: Forgejo,
}

impl Client {
    /// Build from env. `Ok(None)` (not an error) when neither env var is
    /// set — forge isn't configured on this host, which is a normal,
    /// expected deployment shape (see the module doc comment), not a
    /// failure. `Err` only for a genuine misconfiguration: one of the
    /// two env vars present without the other, the token file unreadable,
    /// or the URL/client failing to build — those ARE failures, because
    /// they mean forge was *meant* to be configured here and isn't
    /// usable, which should be loud rather than silently downgrading to
    /// "no forge access."
    pub fn from_env() -> Result<Option<Self>> {
        let url = std::env::var(URL_ENV).ok();
        let token_file = std::env::var(TOKEN_FILE_ENV).ok();
        let (url, token_file) = match (url, token_file) {
            (None, None) => return Ok(None),
            (Some(url), Some(token_file)) => (url, token_file),
            (Some(_), None) => {
                anyhow::bail!("{URL_ENV} is set but {TOKEN_FILE_ENV} is not — partial forge config")
            }
            (None, Some(_)) => {
                anyhow::bail!("{TOKEN_FILE_ENV} is set but {URL_ENV} is not — partial forge config")
            }
        };
        let token = std::fs::read_to_string(&token_file)
            .with_context(|| format!("read forge token from {token_file}"))?;
        let token = token.trim();
        if token.is_empty() {
            anyhow::bail!("forge token file {token_file} is empty");
        }
        let parsed_url =
            url::Url::parse(&url).with_context(|| format!("parse {URL_ENV} ({url}) as a URL"))?;
        let api = Forgejo::new(Auth::Token(token), parsed_url).context("build forgejo client")?;
        Ok(Some(Self { api }))
    }

    /// 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,
        }
    }

    /// Create `name` inside [`CONFIG_ORG`]. Idempotent: an existing repo
    /// (409/422) is folded into success.
    async fn ensure_org_repo(&self, name: &str) -> Result<()> {
        match self
            .api
            .create_org_repo(CONFIG_ORG, Self::repo_option(name, true))
            .await
        {
            Ok(_) => {
                tracing::info!(%name, "swarm forge: created repo in {CONFIG_ORG}");
                Ok(())
            }
            Err(e) if is_already_exists(&e) => {
                tracing::debug!(%name, "swarm forge: repo already exists");
                Ok(())
            }
            Err(e) => Err(e).with_context(|| format!("create repo {CONFIG_ORG}/{name}")),
        }
    }

    /// Add `user` as a collaborator on `CONFIG_ORG/repo` at `permission`.
    /// Idempotent: re-adding an existing collaborator just updates its
    /// permission (forgejo answers 204 either way; a 201 from older
    /// server versions is tolerated defensively).
    async fn add_collaborator(
        &self,
        repo: &str,
        user: &str,
        permission: AddCollaboratorOptionPermission,
    ) -> Result<()> {
        let res = self
            .api
            .repo_add_collaborator(
                CONFIG_ORG,
                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 {CONFIG_ORG}/{repo}"));
            }
        }
        tracing::debug!(%repo, %user, ?permission, "swarm forge: collaborator set");
        Ok(())
    }

    /// 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 agent (a
    /// write-level collaborator, not in the team) cannot merge its own
    /// PR. Mirrors `hive-c0re::forge::repos::apply_operator_branch_protection`
    /// exactly (same policy, same verify-don't-trust shape): a create
    /// failure is ambiguous (already-exists vs. a silent reject that
    /// created no rule), so on any error this GETs the `main` rule and
    /// only treats it as success if the rule is actually present — a
    /// fail-open merge gate is a security bug, not a shrug.
    ///
    /// Also push-whitelists [`SWARM_CONTROLLER_FORGE_USER`] alone —
    /// [`Client::seed_agent_config`]'s initial commit is a direct push to
    /// this same protected `main`, and being a site admin does not
    /// exempt the content-edit API path from a protected branch's push
    /// check the way a real `git push` is exempted (confirmed live: an
    /// already-admin-promoted account still hit `ErrUserCannotCommit`
    /// here). `hive-c0re`'s equivalent seed avoids the whole question by
    /// seeding through an actual `git push`, which the admin bypass does
    /// cover — this daemon has no `git` shell-out, so an explicit
    /// whitelist entry is the narrower fix. The agent (a write
    /// collaborator, not on this whitelist) still cannot push directly,
    /// so this doesn't loosen the "can't merge your own PR" guarantee.
    ///
    /// Assumes [`OPERATORS_TEAM`] already exists in [`CONFIG_ORG`]
    /// (provisioned by `hive-c0re::forge::repos::ensure_operators_team`
    /// on its own startup sweep, not re-provisioned here) — if it
    /// doesn't yet, this fails loudly rather than silently leaving the
    /// repo unprotected, which is the correct failure mode for a
    /// prerequisite that's supposed to already be there.
    async fn apply_operator_branch_protection(&self, repo: &str) -> Result<()> {
        let rule = CreateBranchProtectionOption {
            apply_to_admins: None,
            approvals_whitelist_teams: Some(vec![OPERATORS_TEAM.to_owned()]),
            approvals_whitelist_username: None,
            block_on_official_review_requests: Some(true),
            block_on_outdated_branch: None,
            block_on_rejected_reviews: None,
            branch_name: Some("main".to_owned()),
            dismiss_stale_approvals: None,
            enable_approvals_whitelist: Some(true),
            enable_merge_whitelist: Some(true),
            enable_push: Some(true),
            enable_push_whitelist: Some(true),
            enable_status_check: None,
            ignore_stale_approvals: None,
            merge_whitelist_teams: Some(vec![OPERATORS_TEAM.to_owned()]),
            merge_whitelist_usernames: None,
            protected_file_patterns: None,
            push_whitelist_deploy_keys: None,
            push_whitelist_teams: None,
            push_whitelist_usernames: Some(vec![SWARM_CONTROLLER_FORGE_USER.to_owned()]),
            require_signed_commits: None,
            required_approvals: Some(1),
            rule_name: None,
            status_check_contexts: None,
            unprotected_file_patterns: None,
        };
        let Err(create_err) = self
            .api
            .repo_create_branch_protection(CONFIG_ORG, repo, rule)
            .await
        else {
            tracing::info!(%repo, "swarm forge: applied operator branch protection");
            return Ok(());
        };
        match self
            .api
            .repo_get_branch_protection(CONFIG_ORG, repo, "main")
            .await
        {
            Ok(_) => {
                tracing::debug!(
                    %repo, create_error = %create_err,
                    "swarm forge: operator branch protection already present"
                );
                Ok(())
            }
            Err(check_err) => anyhow::bail!(
                "branch protection for {CONFIG_ORG}/{repo} not applied: create failed \
                 ({create_err}); GET main rule failed ({check_err}), no `main` rule present"
            ),
        }
    }

    /// Create `repo` in [`CONFIG_ORG`] and apply the operator merge gate to
    /// its default branch — the whole job of the `CreateRepo` node.
    /// Branch protection is folded in here rather than a separate node: it
    /// has no independent retry value apart from the repo existing (there
    /// is nothing useful to retry "branch protection on a repo that
    /// doesn't exist yet"), unlike collaborator-add and config-seed, which
    /// are genuinely separate operations against an already-existing repo.
    /// Idempotent — safe to call again.
    pub async fn create_repo(&self, repo: &str) -> Result<String> {
        self.ensure_org_repo(repo).await?;
        self.apply_operator_branch_protection(repo).await?;
        tracing::info!(%repo, "swarm forge: created repo in {CONFIG_ORG} with operator merge gate");
        Ok(format!("{CONFIG_ORG}/{repo}"))
    }

    /// Add `agent` as a **write** collaborator on `repo` (can push + open
    /// PRs, cannot bypass branch protection) — the whole job of the
    /// `AddRepoMember` node. Idempotent.
    pub async fn add_repo_member(&self, repo: &str, agent: &str) -> Result<()> {
        self.add_collaborator(repo, agent, AddCollaboratorOptionPermission::Write)
            .await
    }

    /// Ensure `agent` exists as a Forgejo user account — the whole job of
    /// the `CreateForgeUser` node, and the fix for the "user does not
    /// exist" failure `AddRepoMember` hit before this node existed: adding
    /// a nonexistent user as a collaborator is a Forgejo validation error,
    /// not an idempotent no-op, so something has to create the account
    /// first. Mirrors `hive-c0re::forge::users::ensure_user_exists`'s
    /// intent (an agent's Forgejo identity is provisioned once, up front,
    /// authenticates by token thereafter, and its password is never read)
    /// but not its mechanism: that function shells out to the local
    /// `forgejo admin` CLI, which assumes co-location with the forge host.
    /// This daemon has no such assumption — like every other call in this
    /// file, it only ever talks to the forge over HTTP — so this goes
    /// through `admin_create_user` instead.
    ///
    /// The password itself is a throwaway: 32 random bytes, generated once,
    /// never persisted anywhere, and never needed again (unlike
    /// `hive-c0re`'s CLI path, which can ask forgejo to `--random-password`
    /// on its own, the HTTP admin API requires a real value up front — see
    /// [`crate::webhook::generate_hex_secret`], reused here rather than
    /// duplicated for the same reason a webhook secret and this password
    /// are both "32 random bytes nothing reads back").
    ///
    /// Idempotent: an existing user (409/422) is folded into success, same
    /// as [`Self::ensure_org_repo`]. Deliberately does not attempt to align
    /// the account's email or disable its own repo-creation rights the way
    /// `hive-c0re`'s per-hive provisioning does (`ensure_user_email`,
    /// `ensure_repo_creation_disabled`) — this account only exists so
    /// `AddRepoMember` has something to add, and the agent's owning hive
    /// still runs its own full provisioning pass once the agent actually
    /// spawns there, which self-heals both of those.
    pub async fn ensure_agent_user(&self, agent: &str) -> Result<()> {
        let password = crate::webhook::generate_hex_secret()
            .context("generating a throwaway password for the agent's forge account")?;
        let res = self
            .api
            .admin_create_user(CreateUserOption {
                created_at: None,
                email: format!("{agent}@hyperhive.local"),
                full_name: None,
                login_name: None,
                must_change_password: Some(false),
                password: Some(password),
                restricted: None,
                send_notify: None,
                source_id: None,
                username: agent.to_owned(),
                visibility: None,
            })
            .await;
        match res {
            Ok(_) => {
                tracing::info!(%agent, "swarm forge: created agent forge user");
                Ok(())
            }
            Err(e) if is_already_exists(&e) => {
                tracing::debug!(%agent, "swarm forge: agent forge user already exists");
                Ok(())
            }
            Err(e) => Err(e).with_context(|| format!("create forge user {agent}")),
        }
    }

    /// Seed `repo` with the two files every agent config repo needs:
    /// `agent.nix` (the agent's own module) and `flake.nix` (the
    /// boilerplate that lets the meta flake import this repo as a flake
    /// input) — committed here in one atomic `repo_change_files` call
    /// instead of a local `git commit` (this process has no working tree to
    /// commit from — it only ever talks to the forge over HTTP). The whole
    /// job of the `InitAgentConfigRepo` node.
    ///
    /// **This is the agent's config, not a copy of it.** A hive told to
    /// deploy the agent clones this repo
    /// (`hive-c0re::forge::clone_config_into_proposed`) rather than writing
    /// the same files again locally; the byte-identical template in
    /// `hive-c0re::lifecycle::setup::seed_template` is only reached when no
    /// repo exists here to clone.
    ///
    /// Idempotent by construction rather than by catching a conflict: an
    /// unconditional `repo_change_files` against an already-seeded repo
    /// would overwrite `agent.nix` with the seed content again, clobbering
    /// any edits merged since — so this checks for `agent.nix`'s presence
    /// first and treats it as "already seeded, nothing to do" rather than
    /// re-committing over it.
    pub async fn seed_agent_config(&self, repo: &str, agent: &str) -> Result<()> {
        if self.file_exists(repo, "agent.nix").await? {
            tracing::debug!(%repo, "swarm forge: config already seeded");
            return Ok(());
        }
        let files = vec![
            ChangeFileOperation {
                content: Some(base64_encode(&initial_agent_nix(agent))),
                from_path: None,
                operation: ChangeFileOperationOperation::Create,
                path: "agent.nix".to_owned(),
                sha: None,
            },
            ChangeFileOperation {
                content: Some(base64_encode(initial_flake_nix())),
                from_path: None,
                operation: ChangeFileOperationOperation::Create,
                path: "flake.nix".to_owned(),
                sha: None,
            },
        ];
        self.api
            .repo_change_files(
                CONFIG_ORG,
                repo,
                ChangeFilesOptions {
                    author: None,
                    branch: None,
                    committer: None,
                    dates: None,
                    files,
                    force_overwrite_new_branch: None,
                    message: Some("swarm-controller: seed agent config".to_owned()),
                    new_branch: None,
                    signoff: None,
                },
            )
            .await
            .with_context(|| format!("seed config files in {CONFIG_ORG}/{repo}"))?;
        tracing::info!(%repo, %agent, "swarm forge: seeded agent.nix + flake.nix");
        Ok(())
    }

    /// Whether `path` exists on `repo`'s default branch — the idempotency
    /// check [`Self::seed_agent_config`] uses instead of catching a
    /// create-conflict, since `repo_change_files` has no
    /// already-exists-is-fine status the way `create_org_repo` does.
    ///
    /// `create_repo` makes `repo` with `auto_init: false` (see
    /// [`Self::repo_option`]) — no commits, no `main` ref, nothing a
    /// contents lookup can resolve. Forgejo doesn't 404 that the way it
    /// does a missing path on a real ref: it returns `200` with a bare
    /// JSON `[]`, which fails to deserialize into the single-object
    /// `ContentsResponse` the crate expects for a file path ("invalid
    /// length 0, expected struct `ContentsResponse` with 16 elements" —
    /// serde reading the 0-element array as a positional struct). So a
    /// freshly created, not-yet-seeded repo trips this as an opaque
    /// deserialize error, not the `NOT_FOUND` branch below — confirmed
    /// against a live re-test of the swarm agent-create flow tracked on
    /// the forge. Short-circuit on `empty` first: an empty repo
    /// obviously has no `agent.nix`, and skipping
    /// the contents call for it sidesteps the malformed-response shape
    /// entirely rather than trying to special-case parsing it.
    async fn file_exists(&self, repo: &str, path: &str) -> Result<bool> {
        let is_empty = self
            .api
            .repo_get(CONFIG_ORG, repo)
            .await
            .with_context(|| format!("get repo {CONFIG_ORG}/{repo}"))?
            .empty
            .unwrap_or(false);
        if is_empty {
            return Ok(false);
        }
        match self
            .api
            .repo_get_contents(CONFIG_ORG, repo, path, RepoGetContentsQuery::default())
            .await
        {
            Ok(_) => Ok(true),
            Err(ForgejoError::UnexpectedStatusCode(s)) if s == StatusCode::NOT_FOUND => Ok(false),
            Err(e) => Err(e).with_context(|| format!("check for {path} in {CONFIG_ORG}/{repo}")),
        }
    }

    /// Every agent in [`CONFIG_ORG`] with an open config PR, keyed by agent
    /// name. Mirrors `hive-c0re::forge::config_pr_poll::poll_open_config_prs`'s
    /// scan shape (list repos in the org, list open PRs per repo) but returns
    /// data instead of side-effecting an approval queue — this daemon has no
    /// approval system of its own; it exists so `GET
    /// /api/agents/{name}/config-pr` has something to answer from,
    /// independent of any one hive being up.
    ///
    /// A single repo's list failing does not fail the whole scan — logged and
    /// skipped, so one flaky repo can't blank out every other agent's status.
    pub async fn list_open_config_prs(
        &self,
    ) -> Result<std::collections::HashMap<String, ConfigPrStatus>> {
        let repos = self
            .api
            .org_list_repos(CONFIG_ORG)
            .all()
            .await
            .with_context(|| format!("list repos in {CONFIG_ORG}"))?;

        let mut out = std::collections::HashMap::new();
        for repo in repos {
            let Some(agent) = repo.name.as_deref() else {
                continue;
            };
            let query = RepoListPullRequestsQuery {
                state: Some(RepoListPullRequestsQueryState::Open),
                sort: None,
                milestone: None,
                labels: None,
                poster: None,
                base: None,
                head: None,
            };
            let prs = match self
                .api
                .repo_list_pull_requests(CONFIG_ORG, agent, query)
                .all()
                .await
            {
                Ok(prs) => prs,
                Err(e) => {
                    tracing::debug!(%agent, error = %e, "swarm forge: listing config PRs failed, skipping repo");
                    continue;
                }
            };
            // Only the first open PR matters for the panel — a config repo
            // is meant to carry at most one live proposal at a time (the
            // same assumption `hive-c0re`'s poller and the `MergeConfigPr`
            // approval flow both make).
            if let Some(pr) = prs.into_iter().next() {
                let Some(pr_number) = pr.number.and_then(|n| u64::try_from(n).ok()) else {
                    continue;
                };
                out.insert(
                    agent.to_owned(),
                    ConfigPrStatus {
                        pr_number,
                        html_url: pr.html_url.map(|u| u.to_string()),
                    },
                );
            }
        }
        Ok(out)
    }

    /// Repos with at least one open issue, as `owner/name` full names —
    /// the data source for swarm-ui's repo-filter dropdown, which lists
    /// only repos that currently have open issues rather than every repo
    /// the forge hosts. Filtered on `Repository::open_issues_count` from
    /// the search result itself rather than a follow-up
    /// `issue_list_issues` call per repo — forgejo's own repo summary
    /// already carries that count, so this stays one request regardless
    /// of how many repos exist.
    ///
    /// One search page (forgejo's own default page size) — this binding's
    /// `RepoSearchQuery` has no `page`/`limit` field to page through, unlike
    /// the `(Headers, Vec<T>)`-shaped list endpoints elsewhere in this file
    /// that `.all()` fully drains. Fine for admin-facing tooling at today's
    /// repo count; revisit if an instance's repo count ever outgrows one
    /// page.
    pub async fn list_repos_with_open_issues(&self) -> Result<Vec<String>> {
        let results = self
            .api
            .repo_search(RepoSearchQuery::default())
            .await
            .context("search repos")?;
        Ok(results
            .data
            .unwrap_or_default()
            .into_iter()
            .filter(|r| r.open_issues_count.unwrap_or(0) > 0)
            .filter_map(|r| r.full_name)
            .collect())
    }

    /// Every open issue in `owner/repo`, each with its dependency-derived
    /// `blocked` / `depended_on_by_count` pre-resolved server-side — the
    /// whole job behind `GET /api/repos/{org}/{repo}/issue-report`. Doing
    /// this resolution client-side (one dependency lookup per issue, on
    /// every page load) doesn't scale, which is the reason this endpoint
    /// exists at all rather than swarm-ui resolving it itself per row.
    ///
    /// Forgejo has no bulk or reverse dependency query — the same
    /// limitation `hive-forge issue dependency` hits — so this is still one
    /// `issue_list_issue_dependencies` call per open issue, just run with
    /// bounded concurrency (`CONCURRENCY`) rather than one-at-a-time or all
    /// at once, so a repo with hundreds of open issues doesn't serialize on
    /// round-trips or fire them all in one burst.
    ///
    /// `depended_on_by_count` is a reverse index built off that SAME
    /// resolve pass, not a second query: every open issue's own dependency
    /// list is already in hand to compute `blocked`, so counting how often
    /// each open issue's number appears across all of them is pure
    /// in-memory aggregation. A dependency on a closed issue, or one
    /// outside this open-issue set, can never accrue a count here — it
    /// isn't a row in the report either, so "how many rows depend on this
    /// row" would have nothing to point at.
    ///
    /// `transitively_blocks_count` walks the same `reverse_adj` (blocker
    /// number → issues naming it as a dependency) from each issue — see
    /// [`transitive_reach`] for how the walk itself is made
    /// diamond/cycle-safe. `depended_on_by_count` is `reverse_adj[n].len()`;
    /// this is everything reachable from `n`, not just the direct edge.
    pub async fn issue_report(&self, owner: &str, repo: &str) -> Result<Vec<IssueReportRow>> {
        const CONCURRENCY: usize = 8;

        let issues = self
            .api
            .issue_list_issues(
                owner,
                repo,
                IssueListIssuesQuery {
                    state: Some(IssueListIssuesQueryState::Open),
                    r#type: Some(IssueListIssuesQueryType::Issues),
                    ..Default::default()
                },
            )
            .all()
            .await
            .with_context(|| format!("list open issues in {owner}/{repo}"))?;

        // Numbers collected into an owned `Vec` first, rather than handing
        // `stream::iter` a lazy `filter_map` over borrowed `issues`
        // directly — the latter shape trips a "closure implementation of
        // `FnOnce` is not general enough" HRTB inference error where this
        // function is used as an axum handler (the `routes!` macro checks
        // `Handler` for arbitrary request lifetimes), even though the
        // closure itself is unremarkable.
        let numbers: Vec<i64> = issues.iter().filter_map(|i| i.number).collect();
        let deps: HashMap<i64, Vec<forgejo_api::structs::Issue>> =
            futures_util::stream::iter(numbers)
                .map(|number| async move {
                    let blockers = self
                        .api
                        .issue_list_issue_dependencies(owner, repo, number)
                        .await
                        .with_context(|| format!("list dependencies of {owner}/{repo}#{number}"))?;
                    Ok::<_, anyhow::Error>((number, blockers))
                })
                .buffer_unordered(CONCURRENCY)
                .try_collect()
                .await?;

        let open_numbers: HashSet<i64> = issues.iter().filter_map(|i| i.number).collect();
        let mut reverse_adj: HashMap<i64, Vec<i64>> = HashMap::new();
        for (&number, blockers) in &deps {
            for blocker in blockers {
                if let Some(n) = blocker.number
                    && open_numbers.contains(&n)
                {
                    reverse_adj.entry(n).or_default().push(number);
                }
            }
        }

        let repo_full_name = format!("{owner}/{repo}");
        Ok(issues
            .into_iter()
            .filter_map(|issue| {
                let number = issue.number?;
                let blocked = deps
                    .get(&number)
                    .is_some_and(|b| b.iter().any(|d| d.state == Some(StateType::Open)));
                Some(IssueReportRow {
                    repo: repo_full_name.clone(),
                    number,
                    title: issue.title.unwrap_or_default(),
                    labels: issue
                        .labels
                        .unwrap_or_default()
                        .into_iter()
                        .filter_map(|l| l.name)
                        .collect(),
                    assignees: issue
                        .assignees
                        .unwrap_or_default()
                        .into_iter()
                        .filter_map(|a| a.login)
                        .collect(),
                    html_url: issue.html_url.map(|u| u.to_string()),
                    blocked,
                    depended_on_by_count: u32::try_from(
                        reverse_adj.get(&number).map_or(0, Vec::len),
                    )
                    .unwrap_or(u32::MAX),
                    transitively_blocks_count: u32::try_from(transitive_reach(
                        number,
                        &reverse_adj,
                    ))
                    .unwrap_or(u32::MAX),
                })
            })
            .collect())
    }

    /// [`Self::issue_report`], fanned out over every repo
    /// [`Self::list_repos_with_open_issues`] finds — the default,
    /// no-repo-filter view. Each repo's report is resolved independently and
    /// concurrently (same `CONCURRENCY` bound as the per-issue dependency
    /// resolution inside `issue_report`, since this is the same
    /// "N independent forge round-trips" shape one level up), then
    /// flattened into one row set — `IssueReportRow::repo` is what lets a
    /// caller tell which repo a given row came from.
    ///
    /// One repo's report failing does not fail the whole scan — logged and
    /// skipped, same best-effort shape [`Self::list_open_config_prs`] uses:
    /// a flaky repo should not blank out every other repo's rows.
    pub async fn issue_report_all(&self) -> Result<Vec<IssueReportRow>> {
        const CONCURRENCY: usize = 4;

        let repos = self.list_repos_with_open_issues().await?;
        let rows: Vec<Vec<IssueReportRow>> = futures_util::stream::iter(repos)
            .map(|full_name| async move {
                let Some((owner, repo)) = full_name.split_once('/') else {
                    tracing::warn!(%full_name, "issue_report_all: repo full_name has no '/'");
                    return Vec::new();
                };
                match self.issue_report(owner, repo).await {
                    Ok(rows) => rows,
                    Err(e) => {
                        tracing::warn!(
                            error = %format!("{e:#}"),
                            %full_name,
                            "issue_report_all: one repo's report failed, skipping it"
                        );
                        Vec::new()
                    }
                }
            })
            .buffer_unordered(CONCURRENCY)
            .collect()
            .await;
        Ok(rows.into_iter().flatten().collect())
    }

    /// Register the swarm-wide hooks against this controller, so a real
    /// forge event reaches [`crate::webhook`] instead of the endpoint only
    /// being reachable by hand.
    ///
    /// **These are registered ALONGSIDE the per-hive hooks, not instead of
    /// them.** Every hive keeps receiving and acting on its own deliveries
    /// exactly as today; the controller receives a copy and (for now) logs
    /// it. Taking the hive-side registration away is a later step, and it
    /// has to be later: fan-out swarm→hive does not exist yet, so a hook
    /// moved now would point at a receiver that forwards nowhere — silent on
    /// both sides, indistinguishable from no activity.
    ///
    /// ⛔ **No stale-hook deletion arm, unlike the two per-hive registrars
    /// this otherwise mirrors.** Their arm deletes hooks matching their own
    /// path with a foreign base; copying it here would delete the hives'
    /// live hooks, which are not stale — they are the path still in
    /// production. The controller only ever adds its own.
    ///
    /// Idempotent: an existing hook with the same `target_url` is left
    /// alone, so this is safe on every boot.
    ///
    /// # Errors
    ///
    /// Returns the first failure. A listing failure is not fatal — it falls
    /// through to the create attempt, which is idempotent server-side by way
    /// of the already-exists fold, the same best-effort shape `hive-c0re`
    /// uses.
    pub async fn ensure_swarm_webhooks(&self, public_base: &str, secret: &str) -> Result<()> {
        for kind in DeliveryKind::ALL {
            let target_url = kind.target_url(public_base);
            let (event, scope) = match kind {
                DeliveryKind::Knowledge => (
                    "push",
                    HookScope::Repo {
                        org: KNOWLEDGE_ORG,
                        repo: KNOWLEDGE_REPO,
                    },
                ),
                DeliveryKind::ConfigPr => ("pull_request", HookScope::Org { org: CONFIG_ORG }),
                // Instance-wide: commit/push activity is not scoped to one
                // org (unlike the two above) — it needs observing across
                // every repo the forge hosts, per the "no metric exists for
                // commits or pushes" tracker issue. Forgejo's
                // "global (system) webhook" (`admin_create_hook`) is exactly
                // this — the only scope in this API that fires for a repo
                // in an org created after this hook was registered, with no
                // per-org registration to keep in sync as orgs come and go.
                DeliveryKind::VcsActivity => ("push", HookScope::Instance),
            };
            self.ensure_hook(&scope, &target_url, event, secret)
                .await
                .with_context(|| format!("register swarm webhook {target_url}"))?;
        }
        Ok(())
    }

    /// One hook, idempotently. Split out of [`Self::ensure_swarm_webhooks`]
    /// so the org-scoped and repo-scoped calls do not each grow their own
    /// copy of the list-then-create logic.
    async fn ensure_hook(
        &self,
        scope: &HookScope<'_>,
        target_url: &str,
        event: &str,
        secret: &str,
    ) -> Result<()> {
        // ⚠️ Both non-matching arms log at a level the journal keeps, because
        // this idempotency check fails *silently*: when the list step does
        // not recognise a hook that is already there, the create below runs
        // again, and that stays harmless only while the forge rejects the
        // duplicate. A forge that accepts it leaves one extra hook per
        // process start, each delivering on every event — and the create
        // arm's line reads exactly like a first, correct registration.
        //
        // The two ways to reach a create have different causes, so they are
        // separated: `listed=0` means the create is not landing in the set
        // the list reads (a bucket or permission problem — see
        // [`HookScope::extra_create_config`], which is where the instance
        // scope's own version of that went wrong), a non-zero count with no
        // match means the recorded url is not the one compared.
        match scope.list_hook_urls(&self.api).await {
            Ok(urls) => {
                if urls.iter().any(|u| u == target_url) {
                    tracing::debug!(%target_url, "swarm forge: webhook already registered");
                    return Ok(());
                }
                tracing::info!(
                    %target_url,
                    listed = urls.len(),
                    "swarm forge: no listed hook matches; registering"
                );
            }
            Err(e) => {
                // Best-effort, same as the per-hive registrars: a forge that
                // cannot be listed may still accept a create, and a
                // duplicate create is folded into success below. `warn!`
                // because that fold is an assumption about the forge, not a
                // guarantee — see the note above.
                tracing::warn!(error = %e, %target_url, "swarm forge: listing hooks failed; registering blind");
            }
        }

        let mut additional = BTreeMap::new();
        additional.insert("secret".to_owned(), secret.to_owned());
        for (key, value) in scope.extra_create_config() {
            additional.insert((*key).to_owned(), (*value).to_owned());
        }
        let hook = CreateHookOption {
            active: Some(true),
            authorization_header: None,
            branch_filter: None,
            config: CreateHookOptionConfig {
                content_type: "json".to_owned(),
                url: url::Url::parse(target_url)
                    .with_context(|| format!("parse webhook target url {target_url}"))?,
                additional,
            },
            events: Some(vec![event.to_owned()]),
            r#type: CreateHookOptionType::Forgejo,
        };
        match scope.create_hook(&self.api, hook).await {
            Ok(()) => {
                tracing::info!(%target_url, %event, "swarm forge: webhook registered");
                Ok(())
            }
            Err(e) if is_already_exists(&e) => {
                tracing::debug!(%target_url, "swarm forge: webhook already present");
                Ok(())
            }
            Err(e) => Err(e.into()),
        }
    }
}

/// Every distinct node reachable from `start` by following `adj` edges
/// (`adj[n]` = the issues that directly depend on `n`), not counting
/// `start` itself. Plain DFS with a `visited` set doubling as the
/// cycle guard — see [`Client::issue_report`]'s doc comment for why a
/// cycle has to be tolerated rather than assumed impossible.
fn transitive_reach(start: i64, adj: &HashMap<i64, Vec<i64>>) -> usize {
    // `start` goes into `visited` up front, pre-empting the one case a
    // plain "insert on visit" walk gets wrong: a cycle that loops back to
    // `start` would otherwise re-insert it and count it as its own
    // descendant. Pre-seeding makes that re-visit a no-op instead, so the
    // `- 1` below only ever backs out the seed, never a real node.
    let mut visited: HashSet<i64> = HashSet::from([start]);
    let mut stack: Vec<i64> = adj.get(&start).cloned().unwrap_or_default();
    while let Some(node) = stack.pop() {
        if visited.insert(node)
            && let Some(next) = adj.get(&node)
        {
            stack.extend(next);
        }
    }
    visited.len() - 1
}

/// Where a hook lives. The knowledge hook is repo-scoped and the config-PR
/// hook is org-scoped, mirroring exactly where the per-hive registrars put
/// theirs — a hook on the wrong scope would never fire, and forgejo would
/// report that as a perfectly healthy hook with no deliveries.
enum HookScope<'a> {
    Repo {
        org: &'a str,
        repo: &'a str,
    },
    Org {
        org: &'a str,
    },
    /// Forgejo's "global (system) webhook" — fires for every repo in the
    /// instance, in every org, present or future. The `admin_*` API
    /// namespace; needs the same `write:admin` scope
    /// `Client::ensure_agent_user` already requires on this token, **and**
    /// the `is_system_webhook` config key from
    /// [`Self::extra_create_config`] — that endpoint's default is the
    /// *other* kind of admin hook.
    Instance,
}

impl HookScope<'_> {
    /// Config-map keys the create call needs beyond the shared ones.
    ///
    /// Instance scope carries `is_system_webhook`, and it is load-bearing
    /// twice over. `POST /admin/hooks` reads that key **out of the config
    /// map** and defaults it to `false`, which makes a forgejo *default*
    /// webhook — a template copied into repos created later, not a live
    /// hook — while `GET /admin/hooks` returns only webhooks with the flag
    /// set. Omitting it therefore breaks both halves at once: the hook is
    /// not the instance-wide one this scope exists for, and
    /// [`Self::list_hook_urls`] can never see it, so every process start
    /// creates another one — twelve in a day, `listed=0` each time,
    /// before this key was sent.
    fn extra_create_config(&self) -> &'static [(&'static str, &'static str)] {
        match self {
            Self::Repo { .. } | Self::Org { .. } => &[],
            Self::Instance => &[("is_system_webhook", "true")],
        }
    }

    /// The `url` config value of every hook currently on this scope.
    async fn list_hook_urls(&self, api: &Forgejo) -> Result<Vec<String>, ForgejoError> {
        let hooks = match self {
            Self::Repo { org, repo } => api.repo_list_hooks(org, repo).all().await?,
            Self::Org { org } => api.org_list_hooks(org).send().await?,
            Self::Instance => api.admin_list_hooks().await?,
        };
        Ok(hooks
            .iter()
            .filter_map(|h| h.config.as_ref()?.get("url").cloned())
            .collect())
    }

    async fn create_hook(&self, api: &Forgejo, hook: CreateHookOption) -> Result<(), ForgejoError> {
        match self {
            Self::Repo { org, repo } => api.repo_create_hook(org, repo, hook).await.map(drop),
            Self::Org { org } => api.org_create_hook(org, hook).await.map(drop),
            Self::Instance => api.admin_create_hook(hook).await.map(drop),
        }
    }
}

/// Base64-encode `content` for a forgejo content-API call — every file
/// write on this API takes base64, never raw bytes.
fn base64_encode(content: &str) -> String {
    use base64::Engine as _;
    base64::engine::general_purpose::STANDARD.encode(content)
}

/// Per-agent NixOS module seed — same content
/// `hive-c0re::lifecycle::setup::initial_agent_nix` writes at the per-hive
/// level (this process has no access to that function across the crate
/// boundary, and it's three lines — not worth a shared crate for).
///
/// Deliberately does **not** record which hive the agent belongs to, even
/// though `POST /api/agents` now takes one: an agent's config should carry
/// no reference to the hive it runs on. The hive is an address the swarm
/// routes on, not a property of the agent — a
/// config that named its own hive would be a second place stating where
/// the agent lives, free to disagree with the queue that actually
/// delivers to it.
fn initial_agent_nix(name: &str) -> String {
    format!(
        "{{ config, pkgs, lib, ... }}:\n{{\n  # Per-agent overrides for {name}. This is a regular NixOS module\n  # — add packages, services, modules, imports as needed.\n  #\n  # imports = [ ./extra-module.nix ];\n  # environment.systemPackages = with pkgs; [ ];\n}}\n",
    )
}

/// Module-only flake seed, byte-identical to
/// `hive-c0re::lifecycle::setup::initial_flake_nix` — see that function's
/// doc comment for why the shape looks the way it does (`flakeInputs`
/// forwarding, meta-flake consumption). Not shared across the crate
/// boundary for the same reason [`initial_agent_nix`] isn't.
fn initial_flake_nix() -> &'static str {
    "{\n  description = \"hyperhive agent\";\n  inputs = { };\n  outputs =\n    { self, ... }@inputs:\n    {\n      nixosModules.default = {\n        imports = [ ./agent.nix ];\n        _module.args.flakeInputs = builtins.removeAttrs inputs [ \"self\" ];\n      };\n    };\n}\n"
}

/// Whether a create-style call failed because the object already
/// exists. Forgejo signals this as HTTP 409 (conflict) or 422
/// (validation) — match both defensively, same as
/// `hive-c0re::forge::repos::is_already_exists`.
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,
    }
}

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

    // Both scenarios live in ONE test, not two, deliberately: `cargo
    // test` runs tests in parallel by default, and two tests mutating
    // the same process-global env vars would race each other. A single
    // test's body is one thread, so sequential set/clear within it is
    // race-free without needing a cross-test lock.
    #[test]
    fn from_env_reads_both_vars_or_neither() {
        // SAFETY: test-only mutation of env vars this module itself
        // owns (`URL_ENV`/`TOKEN_FILE_ENV`), scoped to this one test
        // function so there's no cross-test race (see comment above).
        unsafe {
            std::env::remove_var(URL_ENV);
            std::env::remove_var(TOKEN_FILE_ENV);
        }
        assert!(Client::from_env().unwrap().is_none());

        unsafe {
            std::env::set_var(URL_ENV, "https://forge.example.invalid");
        }
        // Not `.unwrap_err()`: that needs `Client: Debug` for its panic
        // message, and the wrapped `forgejo_api::Forgejo` doesn't derive
        // it. Match directly instead of adding a `Debug` impl nothing
        // else needs.
        match Client::from_env() {
            Err(e) => assert!(e.to_string().contains(TOKEN_FILE_ENV)),
            Ok(_) => panic!("expected an error for a partial forge config"),
        }

        unsafe {
            std::env::remove_var(URL_ENV);
        }
    }

    #[test]
    fn transitive_reach_walks_a_chain_not_just_the_direct_edge() {
        // 3 depends on 2 depends on 1: reverse_adj is "who depends on me",
        // so 1 -> [2], 2 -> [3]. Fixing 1 eventually frees both 2 and 3.
        let adj = HashMap::from([(1, vec![2]), (2, vec![3])]);
        assert_eq!(transitive_reach(1, &adj), 2);
        assert_eq!(transitive_reach(2, &adj), 1);
        assert_eq!(transitive_reach(3, &adj), 0);
    }

    #[test]
    fn transitive_reach_counts_a_diamond_descendant_once() {
        // 1 -> [2, 3], both 2 and 3 -> [4]: 4 is reachable via two paths
        // but must only be counted once.
        let adj = HashMap::from([(1, vec![2, 3]), (2, vec![4]), (3, vec![4])]);
        assert_eq!(transitive_reach(1, &adj), 3);
    }

    #[test]
    fn transitive_reach_terminates_on_a_cycle() {
        // 1 -> [2] -> [1]: nothing forge-side prevents a dependency cycle,
        // so the walk has to survive one instead of looping forever.
        let adj = HashMap::from([(1, vec![2]), (2, vec![1])]);
        assert_eq!(transitive_reach(1, &adj), 1);
    }
}