[READ] Classify recent failed pipelines: cause + action per pipeline.

The flagship pipeline RCA: pulls the project's recent failed pipelines with their failed jobs and trace tails, classifies each failed job (test-failure / dependency-network / runner-timeout / oom / script-error) from its failure_reason and trace markers, and attaches a cause and a recommended action. Every classification names its matched evidence, not a black-box verdict. Pass 'failed_pipelines' for pure analysis, or a project to pull live.

Args: project: Project id or full path (required unless failed_pipelines given). limit: How many recent failed pipelines to pull (default 10). tail_lines: Trace-tail lines pulled per failed job (default 60). failed_pipelines: Injected rows {id, ref, jobs:[{name, stage, status, failureReason, traceTail}]}; skips the live pull. target: Server target name from config; omit for the default.

Returns dict: {pipelinesEvaluated, classCounts, pipelines:[{pipeline, ref, headlineClass, cause, action, failedJobs:[{job, stage, class, cause, action, evidence}]}], note}.

[READ] Flag offline/stale/paused runners, long-queued jobs, tag saturation.

The flagship capacity RCA: pulls the runner fleet, flags each runner that is offline, stale (no contact for stale_contact_min minutes) or paused, lists jobs queued past queue_sec, and computes per-tag saturation (queued jobs vs online unpaused runners). Every flag carries its numbers. Pass 'runners' / 'queued_jobs' for pure analysis, or a target to pull the fleet live.

Args: stale_contact_min: Minutes since last contact at which a runner is stale. queue_sec: Seconds a job may wait before being flagged (default 300). saturation_ratio: Flagged queued jobs per online runner at which a tag is saturated (default 2.0). runners: Injected rows {id, description, status, paused, online, tags, contactedAt}; skips the live pull. queued_jobs: Injected rows {id, name, queuedDurationSec, createdAt, tags}. target: Server target name from config; omit for the default.

Returns dict: {runnersEvaluated, flaggedRunners, longQueuedJobs, saturatedTags, thresholds, note}.

[READ] Rank projects by storage; estimate reclaimable artifact bytes.

The flagship storage RCA: pulls project storage statistics, ranks projects by repo + artifact bytes, counts expired-but-kept artifacts and artifacts older than old_artifact_days, and estimates the reclaimable bytes per project. Every ranking carries its byte numbers. Pass 'projects' (and optionally 'artifacts_by_project') for pure analysis, or a target to pull live.

Args: old_artifact_days: Age in days past which a kept artifact counts as reclaimable (default 30). limit: How many projects to pull when live (default 100). projects: Injected rows {path, repoBytes, artifactsBytes}; skips the live pull. artifacts_by_project: Injected map {project: [{file, sizeBytes, createdAt, expireAt}]}. target: Server target name from config; omit for the default.

Returns dict: {projectsEvaluated, totalReclaimableBytes, thresholds, projects:[{project, repoBytes, artifactsBytes, totalBytes, expiredButKept, reclaimableBytes, action}], note}.

[READ] Long-open MRs/PRs, inactive branches, protection config gaps.

The flagship hygiene audit: flags open merge/pull requests idle past stale_mr_days, non-default branches with no commit for stale_branch_days, a default branch without protection, and protection rules that allow force-push. Every flag carries its numbers. Pass injected rows for pure analysis, or a project to pull live.

Args: project: Project id or full path (required unless rows are injected). stale_mr_days: Days an open MR may idle before flagging (default 14). stale_branch_days: Days a branch may idle before flagging (default 90). merge_requests: Injected rows {id, title, state, updatedAt, draft}. branches: Injected rows {name, default, protected, lastCommitAt}. protections: Injected rows {branch, allowForcePush}. default_branch: Default-branch name (pulled live when omitted). target: Server target name from config; omit for the default.

Returns dict: {staleMergeRequests, staleBranches, protectionGaps, counts, thresholds, note}.

[READ] A project's artifact inventory: files, sizes, expiry.

Also reports how many artifacts are past their expiry date but still kept (the delete_artifacts candidates). 'totalBytes' / 'expiredButKept' / 'artifactsFound' count EVERY artifact found, so they stay correct when the row list is cut; 'truncated' then says the rows are a subset. On GitLab the inventory is assembled from a bounded scan of recent jobs — when 'jobScanTruncated' is true, older jobs' artifacts are NOT in this result, so treat the totals as a lower bound.

Args: project: Project id or full path ('group/project' / 'owner/repo'). limit: Max artifact rows to return (default 500). target: Server target name from config; omit for the default.

[READ] Recent pipelines/runs for a project, newest first.

Returns {project, pipelines:[...], returned, limit, truncated}. When 'truncated' is true the server had MORE pipelines than were returned — re-run with a higher limit rather than treating the list as complete. Truncation is measured (one extra row is fetched), never guessed.

Args: project: Project id or full path ('group/project' / 'owner/repo'). status: Optional status filter (e.g. failed, success, running, pending). limit: Max rows to return (default 20, max 99). target: Server target name from config; omit for the default.

[READ] One pipeline/run's full detail.

Args: project: Project id or full path. pipeline: Pipeline/run id (from list_pipelines). target: Server target name from config; omit for the default.

[READ] Jobs of one pipeline/run with status + failure reason.

Returns {project, pipeline, jobs:[...], returned, limit, truncated}. When 'truncated' is true this pipeline had more jobs than were returned.

Args: project: Project id or full path. pipeline: Pipeline/run id (from list_pipelines). limit: Max job rows to return (default 100). target: Server target name from config; omit for the default.

[READ] The tail of one job's log/trace — where the failure explains itself.

Returns {project, job, trace, totalLines, tailLines, returned, limit, truncated, charsTruncated}. 'truncated' true means earlier lines were dropped (the trace is only the tail — the FIRST error may be above the window, so raise tail_lines before concluding a root cause); 'charsTruncated' true means the tail itself hit the byte ceiling.

Args: project: Project id or full path. job: Job id (from pipeline_jobs). tail_lines: How many trailing lines to return (default 60). target: Server target name from config; omit for the default.

[READ] Projects/repositories the token can see, with storage numbers.

Returns {projects:[...], returned, limit, truncated}; 'truncated' true means the server had more projects than were returned (measured, not guessed).

Args: search: Optional name filter. limit: Max rows to return (default 50, max 99). target: Server target name from config; omit for the default.

[READ] One project's detail incl. repo/artifact byte counts.

Args: project: Project id or full path ('group/project' / 'owner/repo'). target: Server target name from config; omit for the default.

[READ] Merge/pull requests for a project (default: open ones).

Args: project: Project id or full path ('group/project' / 'owner/repo'). state: opened/open, merged, closed, all (platform word is translated). limit: Max rows to return (default 50, max 99); the result carries 'truncated' when the server had more. target: Server target name from config; omit for the default.

[READ] Branches with last-commit date and protected flag.

Args: project: Project id or full path. limit: Max rows to return (default 100, capped at 99); the result carries 'truncated' when the server had more. target: Server target name from config; omit for the default.

[READ] Branch-protection rules for a project (incl. force-push flags).

Returns {project, protections:[...], returned, limit, truncated}.

Args: project: Project id or full path. limit: Max rows to return (default 100). target: Server target name from config; omit for the default.

[READ] Releases for a project, newest first.

Args: project: Project id or full path. limit: Max rows to return (default 20, max 99); the result carries 'truncated' when the server had more. target: Server target name from config; omit for the default.

[READ] All runners visible to the token, offline/paused first.

GITLAB ONLY. Gitea has no runner-administration API, so a Gitea target raises a teaching error naming the resources that DO exist there. That is a platform-capability answer, not a broken tool: do not retry it, and do not report the server as unreachable — answer from the pipeline/job surface instead.

Returns {runners:[...], returned, limit, truncated}; rows are sorted offline/paused first BEFORE the cut, so the most urgent runners survive.

Args: status: Optional status filter (online, offline, paused, stale). limit: Max rows to return (default 100, capped at 99). target: Server target name from config; omit for the default.

[READ] One runner's full detail (contacted_at, tags, paused, version).

GITLAB ONLY — a Gitea target raises the same teaching error as list_runners.

Args: runner: Runner id (from list_runners). target: Server target name from config; omit for the default.

[READ] CI/CD server version and revision.

Args: target: Server target name from config; omit for the default.

[READ] The token's identity — who the API sees you as (scope probe).

Args: target: Server target name from config; omit for the default.

[READ] One-shot summary: version, token identity, projects, runners.

Resilient — a failing sub-call degrades to a partial summary with an 'errors' list instead of crashing.

Args: target: Server target name from config; omit for the default.

[READ] List recorded, not-yet-applied undo tokens (most recent first).

Each entry names the original tool, the inverse tool that undo_apply would run, and a human note. Use the undoId with undo_apply.

Returns {undos:[...], returned, limit, truncated}. One extra row is asked for, so 'truncated' is measured — a full page is not assumed to be the end.

Each entry carries effectVerified. False means the original write lost its response, so the change it reverses is PROBABLE, not confirmed — check the live state before applying, and do not report the result as a restore of a state that may never have been reached.

Args: limit: Max rows to return (default 50, max 500). target: Unused (undo state is host-local); accepted for CLI uniformity.

[WRITE][risk=medium] Apply a recorded undo by dispatching its inverse tool.

The inverse runs through its own governed tool, so its real risk tier is recorded there. Pass dry_run=True to preview the inverse call without executing it. A token can only be applied once.

Args: undo_id: The undoId from undo_list (or an _undo_id in a write result). dry_run: If True, preview the inverse tool + params without running it. target: Passed through to the inverse tool when it accepts a target.

[WRITE][risk=medium] Retry a failed/canceled pipeline.

Reads the pipeline first so priorState records the status it had before the retry. Irreversible (a retry is a new run) — no undo. Pass dry_run=True to preview.

Args: project: Project id or full path ('group/project' / 'owner/repo'). pipeline: Pipeline id (from list_pipelines). dry_run: If True, preview without retrying. target: Server target name from config; omit for the default.

[WRITE][risk=medium] Cancel a running pipeline.

Reads the pipeline first so priorState records the status it had before the cancel. Irreversible (the stopped run cannot be resumed) — no undo. Pass dry_run=True to preview.

Args: project: Project id or full path. pipeline: Pipeline id (from list_pipelines). dry_run: If True, preview without canceling. target: Server target name from config; omit for the default.

[WRITE][risk=medium] Pause a runner; reversible (undo resumes it).

Reads the runner first so the harness records its prior paused state. A paused runner stops picking up new jobs; running jobs finish. Pass dry_run=True to preview.

Args: runner: Runner id (from list_runners). dry_run: If True, preview without pausing. target: Server target name from config; omit for the default.

[WRITE][risk=medium] Resume a paused runner; reversible (undo pauses it).

Reads the runner first so the harness records its prior paused state. Pass dry_run=True to preview.

Args: runner: Runner id (from list_runners). dry_run: If True, preview without resuming. target: Server target name from config; omit for the default.

[WRITE][risk=high] Delete a project's artifacts (all, or older than N days).

IRREVERSIBLE — reads the artifact inventory first so priorState records the file count and bytes being destroyed; no undo. Pass dry_run=True to preview (reports what would be reclaimed without deleting).

Args: project: Project id or full path. older_than_days: Only delete artifacts created before this many days ago; 0 = the server's bulk delete of all eligible artifacts. dry_run: If True, preview without deleting. target: Server target name from config; omit for the default.

[WRITE][risk=medium] Protect/unprotect a branch; reversible.

Reads the branch's current protection first so the harness records an undo that replays this same tool with the prior settings. Pass dry_run=True to preview.

Args: project: Project id or full path. branch: Branch name (from list_branches). protect: True to protect the branch, False to remove protection. allow_force_push: Whether the protection permits force-push (default False — the safe setting). dry_run: If True, preview without changing. target: Server target name from config; omit for the default.