Shipped Taskmaster v0.14

Discussions
1

hey friends!

just shipped taskmaster v0.14 :rocket:

→ know the cost of your taskmaster calls
→ ollama provider support
→ baseUrl support across roles
→ task complexity score in tasks
→ strong focus on fixes & stability
→ over 9,500⭐ on github

1. introducing cost telemetry across ai commands

→ costs reported across ai providers
→ breaks down input/output token usage
→ calculates cost of ai command
→ data reported on both CLI & MCP

we don’t store this information yet but it will eventually be used to power model leaderboards on our website.

cost-telemetry
cost-telemetry2038×1978 340 KB

2. ollama provider support

knowing the cost of ai commands might make you more sensitive to certain providers

ollama support uses your local ai endpoint to power @taskmasterai ai commands at no cost

→ use any installed model
→ models without tool_use are experimental
→ telemetry will show $0 cost

ollama support
ollama support2148×2212 411 KB

3. baseUrl support

baseUrl support has been added to let you adjust the endpoint for any of the 3 roles

you can adjust this by adding ‘baseUrl’ to any of the roles in .taskmasterconfig

this opens up some support for currently unsupported ai providers like awscloud Azure

baseUrl support
baseUrl support1560×1162 96.7 KB

4. complexity scores in tasks

after parsing a prd into tasks, using analyze-complexity asks ai to score how complex the tasks are and to figure out how many subtasks you need based on their complexity

task complexity scores now appear across task lists, next task, and task details views

task complexity scores in tasks
task complexity scores in tasks1670×944 69.3 KB

5. lots of fixes & polish

big focus on bug fixes across the stack & stability is now at an all-time high

→ fix MCP rough edges
→ fix parse-prd --append & --force
→ fix version number issues
→ fix some error handling
→ removes cache layer
→ default fallback adjustments
→ +++ more fixes

thanks contributors!

we’ve been cooking some next level stuff while delivering this excellent release

taskmaster will continue to improve even faster

but holy moly is the future bright and i’m excited to share what that looks like with you asap

in the meantime, help us cross 10,000 :star: on github

that’s it for now, till next time!

full v0.14 changelog:

retweet the thread on x

→ npm i -g task-master-ai@latest
→ MCP auto-updates
→ join Task Master AI
→ more info http://task-master.dev
→ hug your loved ones

vibe on friends :grey_heart:

4 Likes
2

hello everybody,

you can also edit the taskmaster system to cut out the api calls by editing the mcp tools to give your local agent in cursor a prompt to work with, essentially the same thing or similar to what the receiving model would be requested through the api endpoint. whenever it calls the mcp tool, it should receive this prompt so it understands how to handle the tool. this way its totally free :smiley:

createPRD, expandTask, analyzeComplexity, etc can all be done locally without needing to use anthropic or an external model this way. can always have the tool prompting do things like telling the model to research using bravesearch or the web tool first to get more context on the potential tasks/sub-tasks.

doing this allows you to do neat stuff like link in instructions to prompt your agent to fill out a knowledge graph mcp with file relations, observations, clustering, etc. so when you use nextTask, itll automatically create everything for your KG based on the completed task :slight_smile:

its also fantastic to do it this way if you dont want to be a part of data collection and have a choice in the telemetry

2 Likes
4

if you’re going to edit the mcp tools to still use the agent to make an AI call (consuming tool use costs or tokens) only to send in a second prompt, that’s going to be a lot of waste

the system prompts for each command uses context about other tasks as needed

you can’t expect to rip out the prompts and expect anything close to the same result

just an fyi to reset those expectations :slight_smile:

5

id be open to seeing a transparent comparison if you have an adequate one

seems to work pretty well for me and ive been doing that for at least a month on all my projects. that single mcp tool use still creates the file (prd/updating tasks.json/creating tasks.json/etc), if the agent runs a web search, its still free unless you’re in max mode.

i set up an mcp tool for initProject that handles the start up too, copies all the files over that are necessary, etc. before all the mcp tools were exposed in an older version iirc.

just my two cents to anyone that wants a cost free version and if anyone’s going to customize the system regardless they might as well tinker with these aspects independent of our opinions.

1 Like
6

I love Task Master so much!

One issue I have is tracking subtasks, I wish you were able to make subtasks within subtasks and have Taskmaster track them properly.

Otherwise, I love it.

7

this sounds legit … whats the difference btwn this and running tasks tho for initproj?

8

I use mcp-shrimp-task-manager