haoranc/estimate

/estimate — AI Agent Effort Estimation

Run PERT three-point estimation with METR reliability thresholds and wave planning for one or more tasks.

Usage

/estimate <task description>
/estimate --file <path>
/estimate --issues <numbers> --repo <owner/name>
/estimate --config <path> <task>
/estimate --format json <task>
/estimate --review-mode none <task>
/estimate --title "My Report" <task>
/estimate validate <observation.yaml>
/estimate calibrate

Instructions

When the user invokes this skill, follow these steps:

1. Parse the invocation

Determine which subcommand to run from the shape of the arguments after /estimate:

| Invocation pattern | Subcommand | | ------------------------------- | -------------------------- | | validate <path> — first token is validate, rest is one .yaml/.yml path plus optional --db <path> | agent-estimate validate | | calibrate — first token is calibrate, rest is empty or only --db <path> | agent-estimate calibrate | | anything else | agent-estimate estimate |

Shape-match strictly: if the arguments start with validate or calibrate but the rest does not fit the shapes above, treat the whole argument string as a task description for agent-estimate estimate — e.g. /estimate validate the login flow is an estimation request, not a malformed validate call.

2. Build the CLI command

Global flags (all subcommands)

| Flag | Short | Description | | ----------- | ----- | ----------------------------------------------------------- | | --verbose | -v | Enable debug logging — useful when a command exits non-zero |

For /estimate

Parse these optional flags from user input and pass them through verbatim:

| Flag | Short | Description | | ---------------------- | ----- | ---------------------------------------------------- | | --file <path> | -f | Path to task file (one task per line) | | --config <path> | -c | Path to config YAML with agent definitions | | --format <fmt> | | Output format: markdown (default) or json | | --type <category> | | Task category: coding, brainstorm, research, config, documentation, frontend, app_dev. Omit to auto-detect. | | --review-mode <mode> | | Review overhead tier: none (0 m), standard (15 m, default), complex (25 m), 3-round (35 m) | | --spec-clarity <n> | | Spec-clarity modifier (0.3–1.3) | | --warm-context <n> | | Warm-context modifier (0.3–1.15) | | --agent-fit <n> | | Agent-fit modifier (0.9–1.2) | | --issues <nums> | -i | Comma-separated GitHub issue numbers | | --repo <owner/name> | -r | GitHub repo (required with --issues) | | --title <text> | -t | Report title |

If none of --file, --issues, or a task description is provided, prompt the user:

Please provide a task description, --file <path>, or --issues <nums> --repo <owner/name>.

Task categories (--type, or auto-detected when omitted):

• coding — feature work, bug fixes, tests, refactors (default tiered PERT model).
• brainstorm — pure ideation and design exploration.
• research — audits, investigations, OSS comparisons, citation/source-grounded work. Research-grounded brainstorms (citation, OSS, benchmark, source, or landscape signals) route here instead of the flat brainstorm band.
• config — deploys, infra, CI/CD, runbooks, monitoring, SRE.
• documentation — API docs, guides, README changes, changelogs.
• frontend — UI/page work (content patches use 15/25/40; page builds use 40/60/90).
• app_dev — app shells and desktop/mobile builds (cold generic L-style prior; use modifiers for warm or highly specified work).

For /estimate validate

The argument after validate is the observation YAML file path. Optionally pass --db <path> if provided.

For /estimate calibrate

Optionally pass --db <path> if provided. Default: ~/.agent-estimate/calibration.db.

3. Run the command

Execute the CLI via Bash:

agent-estimate estimate "Add login button"
agent-estimate estimate --file tasks.md
agent-estimate estimate --issues 1,2,3 --repo org/name
agent-estimate estimate --config custom.yaml "Add login button"
agent-estimate estimate --format json "Add login button"
agent-estimate validate observation.yaml
agent-estimate validate observation.yaml --db ~/.agent-estimate/calibration.db
agent-estimate calibrate
agent-estimate calibrate --db path/to/db

Capture stdout and stderr. If the command exits non-zero, display the error message to the user.

4. Display output

Display the CLI output directly to the user. No post-processing — the CLI handles formatting.

Examples

/estimate Add a login page with OAuth
→ agent-estimate estimate "Add a login page with OAuth"

/estimate --file spec.md
→ agent-estimate estimate --file spec.md

/estimate --issues 1,2,3 --repo myorg/myrepo
→ agent-estimate estimate --issues 1,2,3 --repo myorg/myrepo

/estimate --config agents.yaml --format json "Refactor auth module"
→ agent-estimate estimate --config agents.yaml --format json "Refactor auth module"

/estimate validate results/sprint1.yaml
→ agent-estimate validate results/sprint1.yaml

/estimate validate results/sprint1.yaml --db ~/.agent-estimate/calibration.db
→ agent-estimate validate results/sprint1.yaml --db ~/.agent-estimate/calibration.db

/estimate calibrate
→ agent-estimate calibrate

Observation YAML format (for /estimate validate)

task_type: feature
estimated_minutes: 45.0
actual_work_minutes: 52.0
actual_total_minutes: 60.0
file_count: 3
line_count: 120
test_count: 5
execution_mode: single
review_mode: standard
review_overhead_minutes: 8.0
modifiers:
  spec_clarity: 1.0
  warm_context: 0.9

Notes

• Requires agent-estimate installed: pip install agent-estimate or pip install -e .[dev] in the repo.
• Default config uses bundled default_agents.yaml. Pass --config to override agent definitions.
• --review-mode defaults to standard (15 m additive; clean 2x-LGTM). Use complex (25 m) for involved or security-sensitive review, 3-round (35 m) for explicit 3-round cross-agent review, or none for self-merge workflows.
• JSON output is available via --format json.
• METR p80 reliability thresholds back the over-threshold warnings. Current model keys: opus_4_7 (Opus 4.7, 90 m), gpt_5_5 (90 m), gpt_5_4 (60 m), gemini_3_1_pro (45 m), sonnet_4_6 (30 m), haiku_4_5 (15 m). opus_4_x is a forward-compatible Opus alias; legacy keys (opus_4_6, opus, gpt_5/5.2/5.3, gemini_3_pro, sonnet) remain accepted.