time-estimate

name	time-estimate
description	estimate CC active time for a task using the time rule, with dynamic effort resolution
argument-hint	<task description>
user-invocable	false

/time-estimate

produce a realistic CC active-time estimate for the given task, grounded in plugins/cc/rules/time.md. never guesses effort level.

what to do when invoked

1. resolve active effort, highest-precedence first. stop at the first rung that returns a value:

   • echo "${CLAUDE_CODE_EFFORT_LEVEL:-}" (env var)
   • --effort flag, if visible via process context
   • /effort session state, if already set this session
   • effortLevel key in ~/.claude/settings.json, then any project-scoped .claude/settings.json under the repo root
   • otherwise, model default

   cite the rung that supplied the answer. if ambiguous, ask the user rather than guess.

2. resolve active model from the session's model id (e.g. claude-opus-4-7).

3. classify session mode using the table in the rule:

   • quick fix: single-file edit, typo, config tweak
   • standard: feature work, refactor, small module
   • marathon: migration, infra build, multi-file refactor

4. apply the multiplier from the model × effort matrix in the rule. quote the baseline at low, then scale.

5. return the estimate in the rule's format:

   • CC: <range> (always a range, never a single number)
   • effort rung cited (e.g. "rung 4, settings.json")
   • session mode named
   • your-time estimate for review
   • confidence (high / medium / low)
   • risks that could 2× the number

6. if the task breaks into sub-phases, estimate each and sum. flag which phases parallelize via subagents (1.8-2.2×) and which stay serial.

rules to respect

• never quote a single number. always a range.
• never assume xhigh or low without checking.
• always distinguish CC time from your time.
• if on opus 4.7 with heavy work, offer the effort trade: xhigh is ~30% slower than high for throughput-bound work.
• never suggest fast mode on opus 4.7 or sonnet 4.6.

example output shape

task: add logout button with tests

effort: low (rung 4, ~/.claude/settings.json)
model: claude-opus-4-7 (0.87× baseline)
mode: standard

CC: 12-18 min active
your time: ~10 min review
confidence: medium

phases:
  P1 find auth module (serial): 2-3 min
  P2 write button + route (serial): 4-6 min
  P3 tests (parallel via subagent): 3-5 min wall, ~6-9 min work
  P4 verify (serial): 3-4 min

risks that 2×: auth module structure unclear, test framework unfamiliar

notes

the rule's matrix is a generic baseline. for personal calibration, run /time-calibrate.