Loading ToolBench
Local task-state ledger for AI coding agents. Provides durable task continuity, source caching, evidence tracking, and decision recording across agent sessions.
Agentpack demonstrates strong definition quality across 21 well-structured tools. All tools have clear descriptions (100-400 chars average), proper naming conventions (verb-first, context-specific), and complete input schemas with typed parameters. Tool annotations are correctly implemented per MCP 2026-07-28 spec. However, output schemas are not explicitly documented, responses are returned as formatted text or JSON but the structure is not formally described. Error handling is present but somewhat generic; some tools lack specific recovery guidance. The server is conservative in design, reflecting its specialized role as a task-state ledger for agent collaboration.
Store verification output (test results, command output, review findings, notes, or links) as a file under .agentpack/evidence/ plus a ledger event, returning an evidence id to reference from task_update_verification, task_finalize, or record_decision. Call for meaningful verification worth preserving; for small tasks prefer one aggregated evidence item over many per-command items. Provide the body inline via content or from a file via path.
Export a task bundle: task passport, decisions, dead ends, source conclusions, and evidence for migration to another repo's Agentpack or archival. Output is JSON; bundle can be imported via bundle_import. Read-only.
Import a task bundle (decisions, dead ends, source conclusions, evidence) into the current Agentpack. By default resumes the current task with bundled context; use asNew=true to import as a new parked task. Records import event(s) under .agentpack/.
Preview a task bundle import into the current Agentpack: show task title, decisions, dead ends, source conclusions, and evidence, plus any conflicts or overwrite warnings. Read-only; does not commit changes.
Inspect a task bundle (JSON file exported from bundle_export): show what will be imported, including task title, decisions, dead ends, source conclusions, and evidence summaries. Read-only.
Output schemas not formally documented. Tools return formatted text (markdown) or JSON but the response structure is not explicitly declared in a schema. LLMs cannot predict response fields without example-driven inference.
task_park has an empty input schema (no properties). While semantically correct (the action takes no params), the schema is minimal and offers no guidance. Could document the precondition that a task must be active.
Error handling is present but lacks recovery guidance in most tools. Errors indicate what failed but do not suggest next steps (e.g., 'Task not found. Try task_list() to see available tasks.')
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | B | 75 | 2026-07-28+ | v2 |
Load a token-budgeted markdown resume of Agentpack state for the current task: Task Passport status and next actions, git state, query-relevant decisions, dead ends, and source conclusions, plus gate warnings when the task lifecycle needs attention. Call once at the start of a session or task, before reading code; re-call only for a different query or budget. Read-only.
Record an approach that failed so future agents do not repeat it. Call when an attempted direction is abandoned for a durable reason, not for ordinary debugging iterations. Writes one event under .agentpack/; secret-like values are redacted.
Append a durable technical or product decision to the Agentpack ledger so future sessions inherit it. Call for decisions that matter beyond this session (architecture, contracts, tradeoffs), not for routine preferences or per-edit narration. Writes one event under .agentpack/; secret-like values are redacted.
Record a durable conclusion about a source file in the Source Cache: stores the file's current content hash with your summary so future sessions can reuse the conclusion until the file changes. Call after inspecting an important file when the conclusion is reusable; do not record every file read, and re-record only when the conclusion itself changed. Writes under .agentpack/.
Run Agentpack release preflight checks: verify that the task is finalized, decisions are durable, source cache is coherent, evidence is complete, and there are no stale or broken references. Fails with a report of blockers and advisory warnings. Read-only.
Export a token-budgeted markdown resume of Agentpack state for handoff or review: task status, decisions, dead ends, source conclusions, evidence summaries, and next actions. Use budget or preset to control size. Read-only.
Check whether recorded source conclusions are unchanged, changed, or missing by re-hashing the files; use changed/missing filters for stale source-cache triage. Call when you need a full stale-source check beyond what load_context already showed; do not repeat it when a recent load_context, task_audit, or status check answered the question. Read-only.
Audit the current Task Passport for continuity risks and advisory-only risk-proportional adversarial-verification evidence (a concise verification note or test output at low risk; independent read-only review and a named disconfirming check at medium/high risk). It does not judge semantic correctness or block lifecycle actions. Call before finalizing to surface risks that matter to task continuity.
Mark the current task as complete and archive it. Fails if verification is not passed or accepted, or if there are unresolved gate warnings. Call only when the task is truly done. Records one event under .agentpack/.
Full task handoff for resumption by a different agent or session: current task status, decisions, dead ends, source conclusions, evidence, and next actions. Exports a token-budgeted markdown resume plus a task bundle for import into another repo's Agentpack. Read-only; returns markdown and bundle JSON separately.
List all Task Passports: parked, finalized, and current (marked with *). Use json=true for structured parsing. Read-only.
Park the current task: mark it as paused, safe to resume later, and ready for task_switch. The task remains in the ledger and can be resumed or finalized. Records one event under .agentpack/.
Start a new Task Passport (the current task), or fail if one already exists. Always set a clear title and objective. Use --write-scope to declare files this task touches, --next to list planned actions, and --risk to convey task continuity risk (low/medium/high). Records one event under .agentpack/.
Report the current Task Passport: title, objective, write scope, next actions, risk, passport status, and git state. Read-only.
Switch the current task to a parked or finalized task (by id or title prefix). Fails if the target does not exist. Records one event under .agentpack/.
Update Task Passport verification status (passed, failed, accepted) and attach evidence. Call after running verification checks or reviews. Records one event under .agentpack/.
Destructive tools (record_decision, record_dead_end, task_finalize, bundle_import) lack dry-run or confirmation patterns. Agents could inadvertently record incorrect data without a review step.