Feasibility Study

Feasibility Study Skill

Supplementary Agent

For each solution option, dispatch background exploration:

Agent({ description: "Explore feasibility of solution option", subagent_type: "feasibility-analyst", prompt: Research the feasibility of: <solution description> Evaluate technical feasibility, effort, risk, extensibility, and maintenance cost. })

Trigger

• Keywords: feasibility, is this possible, can we, should we, explore options, before tech spec

When NOT to Use

• Already have a tech spec (use /deep-analyze)
• Need implementation, not analysis (use /codex-implement)
• Quick question (use /codex-explain or /codex-architect)

Workflow

Resolve → Decompose → Constraints → Code research → Solutions → Codex discussion → Decision → Report

Phase 0: Resolve the feature context

The sets Phase 1 reads have to come from somewhere, and this skill is invoked directly (/feasibility-study <topic>) as often as it is invoked from another skill — so it resolves them itself rather than assuming a caller supplied them:

# The node entrypoint. The shell wrapper's own header tells skills to prefer this one once they hold
# `Bash(node:*)` — and this skill now does, for its transport dispatches. The wrapper was named here
# only while that grant was absent; keeping it afterwards would have instructed a second-choice
# entrypoint for a reason that had stopped being true.
node scripts/resolve-feature.js [--feature <key>]

The reply is one JSON document. What this skill reads out of it: scan_error first — the gate below decides whether anything else in the payload means what it says — then current_authority (what the system does today) and the design_records entries — design records carry the rationale; it is the type: requirements subset of them that states what was asked for, which is why Phase 1 filters before it selects. Each entry is { file, type, namespace, confidence, is_canonical, role }, file relative to docs_path. An empty or non-JSON reply is a failure too — node may be unavailable, which the shim cannot report as a payload. Treat it exactly as scan_error !== false below.

Phase 1: Requirement Decomposition

Input source priority:

1. If a requirements doc resolves from design_records → consume as the authoritative statement of what was asked for, validate via 5-Why. It is a design record, not a description of current behaviour: for "what does the system do today", read code, rules/ and current_authority
2. Otherwise → extract requirements from user input via 5-Why analysis

design_records is an array, so "a requirements doc" needs a rule rather than an assumption — a split or variant-backed phase contributes more than one. Filter to type: requirements first, then:

Rows 1 and 4 are different answers: "there is none" is acted on, "there are two" must not be resolved by picking. The same order /architecture applies to its tech-spec candidates.

scan_error gate. Gate on scan_error !== false, not on scan_error === true. When it is not exactly false the four source sets are unknown, not empty — the corpus could not be enumerated (unreadable directory, broken taxonomy, no repository), or the resolver never ran and a shell fallback supplied a payload with no such field at all. {} is the shape that made the stricter test useless: it has no scan_error, so === true is false and the gate passes a payload that contains nothing. Do not proceed as though the feature has no authority documents — report and take the ⚠️ Need Human exit. A key may still be present, so a non-null key is not evidence the sets are complete.

Use "5 Why" to uncover essence:

1. Surface requirement (what user asks for)
2. Underlying problem (why they need it)
3. Success criteria (quantifiable acceptance)