TaskflowSAFE
DAG runtime for coding-agent workflows with verification, resume, replay, and effect checks.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
[](https://github.com/heggria/taskflow/actions/workflows/ci.yml) [](https://nodejs.org) [](./LICENSE) [](#host-adapters)
English · 简体中文
1.0 overview · Quickstart · Docs · Examples · Changelog
taskflow is a declarative runtime for coding-agent workflows. It turns a graph into a verifiable execution contract, runs phases in isolation, and keeps intermediate work out of the host conversation. In Taskflow 1.0, the contract also describes the effects a phase is allowed to propose.
Taskflow 1.0.0 includes the DAG runtime, trusted filesystem effects and complete Control Plane (CLI/MCP/WebUI). GitHub Releases and npm establish publication; see the acceptance scoreboard for verified scope and limits.
The declared-effect contract
An agent can propose content. It should not become the mutation authority merely because it can run a command.
For admitted, declared filesystem-write targets, taskflow 1.0 makes the path explicit and routes the final mutation through the resources transaction:
flow / .tf.ts │ ▼ validate + verify ──► EffectIR + FlowIR hash │ │ │ ▼ │
7ec2af7b76b0OBSERVED · 2026-10-08Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.
claude mcp add taskflow-website --env TASKFLOW_E2E_TOKEN=${TASKFLOW_E2E_TOKEN} -- npx -y [email protected]{
"mcpServers": {
"taskflow-website": {
"command": "npx",
"args": [
"-y",
"[email protected]"
],
"env": {
"TASKFLOW_E2E_TOKEN": "${TASKFLOW_E2E_TOKEN}"
}
}
}
}Exposed tools (76)
60 read · 15 write · 1 destructive. Blast radius: 1 tool can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.
| Tool | Risk | Description |
|---|---|---|
a | read | test agent |
alpha | read | first |
analyst | read | built-in analyst |
audit-endpoints | read | audit endpoints for auth |
auditor | read | audits code |
b | read | test agent b |
bad-thinking | read | test |
bare | read | minimal |
beta | read | second |
builder-agent | read | builder |
builtin | read | package agent |
csv | read | csv tools |
deep | read | found deep |
default | read | test |
empty | read | no body |
executor | read | test executor |
full | read | fully specced |
gamma | read | third |
good | read | readable |
gp-agent | read | from grandparent |
judge | read | judge agent |
large | read | x |
linked | read | via symlink |
literal-agent | read | literal |
my-flow | read | desc |
nomodel-agent | read | no model |
ok | read | fine |
p | read | proj |
packed | read | packed |
parent-agent | read | from parent |
pathcheck | read | check |
planner | read | test |
probe | read | Harmless compatibility probe |
reviewer | read | test |
role-agent | read | role |
roundtrip | read | full fidelity |
scout | read | finds things |
scout-agent | read | scout |
spaces | read | spaced tools |
stall | read | d |
t | read | t |
taskflow | read | Taskflow |
taskflow_analytics | read | Summarize the last N runs of a flow: status histogram, p50/p95 duration, per-phase fail and cache-hit rates. Zero tokens, no writes, no auto-tune. |
taskflow_compile | read | Render a flow |
taskflow_control_approval_decide | write | Approve, reject, or edit a pending approval with its expected run version. Edit requires a staged artifact and editKind. |
taskflow_control_approval_stage_edit | write | Stage bounded output or plan text as a ledger-reachable approval edit artifact. Refresh status before deciding. |
taskflow_control_approvals | read | List durable approval requests in the authorized project. |
taskflow_control_cancel | write | Cancel a run through the Host |
taskflow_control_coordinator | read | Read user-global capacity and only this project |
taskflow_control_force_release | destructive | Explicit operator only: release capacity without proving execution stopped; permanently marks operator-overridden. Requires acknowledgement and an audit reason. |
taskflow_control_projects | read | List projects authorized for this local session. |
taskflow_control_receipt | write | Read the authorized evidence receipt for a terminal run. |
taskflow_control_result | write | Read the authorized run result, including final output only. |
taskflow_control_resume | write | Resume a recoverable run through Host admission and authority checks. |
taskflow_control_runs | write | List durable run status in the authorized project. |
taskflow_control_set_max_active_runs | write | Explicit operator only: set the user-global admitted-run limit. Reuse commandId for exact retries. |
taskflow_control_status | read | Read durable status without phase transcripts. |
taskflow_control_submit | write | Submit a flow to the authenticated Host. Reuse commandId when retrying the same request. |
taskflow_control_why | write | Explain authorized durable run evidence and effect causes. |
taskflow_lint | write | Run built-in and project-local pluggable verifiers (script-lint, custom checks) on a flow WITHOUT executing it. Discovers verifiers from .pi/taskflows/verifiers/. Provide |
taskflow_list | read | List the saved taskflows discoverable from the current working directory (user + project scope). |
taskflow_plan | read | Terraform-plan for taskflows: bind typed args, verify structure, project phase order + dynamic/unresolved bindings, and compute a worst-case agent-call bound — WITHOUT spawning any agent. Provide |
taskflow_replay | write | Re-evaluate a stored run |
taskflow_run | write | Run a taskflow DAG and return its final output. Provide EITHER |
taskflow_runs | write | List recent background runs, inspect one run, wait for completion without losing it when the MCP request ends, or request cancellation. Use after taskflow_run with mode: |
taskflow_show | read | Show a saved flow |
taskflow_verify | read | Statically verify a flow (cycles, missing deps, undefined refs, ...) WITHOUT executing it. Provide |
taskflow_version | write | Report the engine package version, the git commit the dist was built from, the bound host identity (codex/claude/opencode/grok/hermes), and the run-state schema version. Zero tokens, no execution. Use to verify which taskflow build a host is running. |
taskflow_why_effect | read | Given a runId + effectId (+ optional phaseId): explain authorization and lifecycle from the durable resource intent ledger. Zero tokens and read-only. A declaration without matching principal/capability/intent evidence is unauthorized. |
taskflow_why_stale | read | Given a runId (+ optional phaseId seed): with no seed, prints the observed dependency graph; with a seed, computes the transitive stale frontier — exactly which phases would need re-running and why (observed ∪ declared edges). Zero tokens. Read-only. |
test | read | |
tooled | read | has tools |
u | read | user |
unk | read | unknown |
worker | read | test worker |
writer | read | test |
Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | PASS |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | WARN |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (1 observation(s))
- Network
- declared (1 observation(s))
- Shell
- declared (1 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- none-found
Findings (17)
taskflow_control_force_release
} from "../../../scripts/verify-published-package.mjs";
const repo = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "../../..");
} from "../../src/project/types.ts";
const repo = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "../../..");
} from "../../../examples/charterarc-phase-docs.ts";
- **Flow hooks (`hooks.onComplete` / `onFail` / `onBlocked`).** Fire-and-forget `webhook` | `file` | `command` (argv only) with summary payload `taskflow.hook.v1` (never transcripts). `https` and `htt
- `url` must be `https:` (or `http://127.0.0.1` for local dev — document)
| 2 | Webhook localhost | **Allow** `https:` and `http://127.0.0.1` / `http://localhost` (other `http:` rejected) |
- Allow: `https://...`, `http://127.0.0.1...`, `http://localhost...` (and `[::1]` if we parse IPv6 loopback)
"url": "http://127.0.0.1:8123/v1/embeddings",
@types/node, typebox, typescript
typescript
typescript
typescript
@types/mdx, zod
Claude Code is a powerful coding agent on its own. But when a job needs a **repeatable pipeline** — review every changed file, gate on risk, and set an observed-usage stop-loss — ad-hoc prompting runs
Gates applied: no_behavioural_pass.
7ec2af7b76b0full audit observations/trust-audit/mcp-server/heggria__taskflow-5.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-08 | 7ec2af7b76b0 | SAFE | B | 89 | first audit |
Questions
What is the Taskflow MCP server?
DAG runtime for coding-agent workflows with verification, resume, replay, and effect checks.
What tools does Taskflow expose?
76 in total: 60 read-only, 15 that write, and 1 that can delete or overwrite (taskflow_control_force_release). Every one is listed on this page with its risk.
Is Taskflow safe to connect to an agent?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean server reads B. Separately from the audit: 1 of its tools can destroy data, so scope the token you give it to what you actually need.
What credentials does Taskflow need?
It reads TASKFLOW_E2E_TOKEN from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.
How current is this page?
The grade is for one exact copy of the source (7ec2af7b76b0), read on 2026-10-08. The repository is watched and re-audited when it changes.