SkillPartly tested (needs an account)

Realistic estimates for agent builds

mythical-agent-month

Estimation skill that splits agent wall-clock, human effort, waiting time and usage limits, so AI stops quoting months of team time for agent-built work.

Works with
Claude Code, Codex CLI, Gemini CLI
Good for
Build an app

What it does

Mythical Agent Month is an estimation skill. Ask an assistant how long a roadmap will take and it often answers in human-team months. This skill makes it split the answer into machine time (agent wall-clock and agent-hours along the critical path, with the parallelism you really get), human effort (prompting, review, setup), waiting (decisions and usage-limit resets) and the share of your weekly usage limit.

It has two modes: you driving an assistant task by task, or a plan split into tasks that agents build on their own. A small Python script does the arithmetic (PERT per task and the critical path). A second script reads your own Claude Code subagent transcripts to replace the default numbers with yours. If a human team will do the work without AI, the skill says it does not apply.

Example prompts

  • "how long will this take?"
  • "estimate this backlog"
  • "how long will this workflow run?"

How to install

/plugin install mythical-agent-month --marketplace MedAdemBHA/mythical-agent-month

How to use it

Ask how long work will take, for example how long will this take? or estimate this backlog. The skill picks developer-plus-AI or autonomous-agent mode, asks at most two questions, and prints a plan number with the likely value and range beside it.

Check it works

After installing, start a new session and ask "how long will this take?" about a small feature. The answer should keep machine time, human effort and waiting time apart and show a bold plan number with a likely value and a range, not a single number of weeks. With Python 3.12 or later you can also run python3 scripts/totals.py ../../examples/agent-feature.json inside skills/mythical-agent-month to see sample output.

Common problems

Other assistants

What to watch out for

  • Plugin route needs Claude Code 2.1.275 or later for the in-session /plugin form (README); the shell form claude plugin install ... --marketplace needs 2.1.292
  • Loads by itself: SKILL.md has no disable-model-invocation, and its description asks for it whenever you ask how long work will take or want a timeline or roadmap sizing
  • While it estimates, the skill tells the assistant to read your code for each task, write the tasks as a JSON file and run scripts/totals.py with Python
  • scripts/totals.py needs Python 3.12 or later. On Python 3.11 it stops with a SyntaxError (line 284), so the macOS built-in python3 cannot run it; the skill then tells the assistant to do the arithmetic by hand. The README says only Python 3
  • install.sh copies the skill folder into each agent's skills folder and first deletes any existing folder with the same name there (rm -rf of <skills>/mythical-agent-month); 'all' writes to Claude Code, Codex, Cursor, Gemini CLI and Copilot folders
  • The README offers a curl | bash one-liner; the manual git clone and cp route does the same thing without piping a script
  • Defaults come from one project (n = 1 by the author's own account); held-out testing was 19% under on the plan number, so treat results as order-of-magnitude
  • calibrate.py reads your Claude Code subagent transcripts under ~/.claude/projects/ (timestamps, labels, model names, token counts) and prints no message content; it makes no network calls. The optional --csv file holds each subagent's label

Licence and author

Licence
MIT
Author
MedAdemBHA
Source code
github.com/MedAdemBHA/mythical-agent-month

How we tested it

We installed it on an empty test computer on 8 Nov 2026. Partly tested: installed in a clean test system, but its main features need an account or key we did not use.

Test level

Not run in a live session. totals.py needs Python 3.12 or later. For the manual route, run mkdir -p ~/.claude/skills first.

Technical details of the test
Date
8 Nov 2026
Method
Reviewed by our team against the repository at b60165f (v1.2.0): README, SKILL.md, install.sh and both scripts. In a clean container (Claude Code 2.1.296) the shell plugin install worked and claude plugin list showed version 1.2.0 enabled. ./install.sh claude codex gemini put the folder in all three skills folders. The README's manual cp line failed because ~/.claude/skills did not exist, and totals.py failed on Python 3.11.

Checked again on 8 Nov 2026

Alternatives

  • No catalog entry does the same job yet; Superpowers writes plans but gives no time or usage estimates

FAQ

  • Q: Does it send my transcripts anywhere? A: The scripts make no network calls; calibrate.py reads metadata locally and prints no message content. Anything the assistant reads while it estimates goes to your AI provider as usual.
  • Q: How accurate is it? A: The author reports order-of-magnitude accuracy from one project and asks for more runs; the printed plan number is the likely value times 1.3.
  • Q: Is it for teams without AI? A: No. SKILL.md tells the assistant to give a normal estimate and say the skill does not apply.

Sources

  • github.com/MedAdemBHA/mythical-agent-month
  • github.com/MedAdemBHA/mythical-agent-month/blob/main/skills/mythical-agent-month/SKILL.md
  • github.com/MedAdemBHA/mythical-agent-month/blob/main/install.sh