Research

Research

Always run this skill after /shape and before /write-a-prd. It auto-calibrates depth: sometimes it's a 2-minute version check that confirms your approach is still valid, sometimes it's a 30-minute deep dive into unfamiliar territory. The point is that it always runs — you don't get to skip it — because the most expensive research failures are the ones where you didn't know you needed to research.

This is also the first anti-anchoring checkpoint in the workflow. If the user or an earlier conversation introduced a date, budget, or confidence claim, clarify whether it is a target, an estimate, or a commitment before carrying it forward into the PRD.

One question per turn. When confirming constraints, reviewing findings, or asking the user anything, ask one question at a time and wait for the answer before asking the next. Never present a batch of questions as a numbered or bulleted list.

Prefer single-select. Use single-select multiple choice when the user is choosing one direction, one priority, or one next step.

Use multi-select rarely. Reserve it for compatible sets — goals, constraints, non-goals, success criteria — that can all coexist. If prioritization matters, follow up asking which selected item is primary.

Use the platform's question tool when available. In Claude Code, use AskUserQuestion; in Codex, request_user_input; in Gemini, ask_user. Otherwise, present numbered options in chat and wait for the user's reply before proceeding.

Invocation Position

This is a primary pipeline skill, not an optional extra. The default flow is /shape → /research → /write-a-prd.

Use /research whenever the work is headed toward shaping or implementation and you need to verify current versions, official docs, existing code patterns, and past solutions before making design commitments.

Do not skip this step just because the approach feels obvious. Do not use it as a replacement for /shape when the problem is still underdefined, and do not use it as a replacement for /execute once the work is already shaped and ready to implement.

Why This Is Mandatory

AI models are trained on internet-scale codebases where deprecated patterns outnumber current ones by orders of magnitude. The model will confidently generate middleware.ts for a Next.js 16 project where the file was renamed to proxy.ts. It will use getServerSideProps in an App Router project. It will suggest packages that don't exist. The only defense is to verify against reality before committing to an approach.

Execution Flow

Phase 0: Environment & Version Check (MANDATORY — never skip)

This phase takes 30-60 seconds and catches the most dangerous class of errors: stale API patterns from training data.

1. Read the dependency manifest. Extract installed versions of all major dependencies:

   # JavaScript/TypeScript projects
   cat package.json | jq '.dependencies, .devDependencies' 2>/dev/null
   
   # Python projects
   cat requirements.txt 2>/dev/null || cat pyproject.toml 2>/dev/null
   
   # Rust projects
   cat Cargo.toml 2>/dev/null
   

2. Identify the key dependencies for this feature. Based on what /shape established, which dependencies will this feature touch? List them with their installed versions.

3. Check for breaking changes. For each key dependency, do a targeted web search:

   "[dependency name] [installed version] breaking changes migration"
   "[dependency name] [installed version] vs [latest version] API changes"
   

   Look specifically for:

   • API renames (middleware.ts → proxy.ts)
   • Removed or deprecated methods
   • Changed function signatures (sync → async)
   • Changed file conventions (pages/ → app/)
   • Changed import paths
   • Young features: features the feature-under-research depends on that the changelog or docs flag as "Added in X.Y" / experimental / beta / alpha, where the gap from X.Y to the installed version is small (rough heuristic: introduced within ~6 months of the installed version's release). Surface these as a 📦 YOUNG FEATURE flag listing the affected feature names — they need an upstream-issue-tracker audit in Phase 4 step 3 before the recommendation is finalized. Docs describe what is supposed to work and types describe what compiles; only the issue tracker describes what is currently broken in version X.Y against integration Z.

4. Check CLAUDE.md / AGENTS.md. Read any existing project context files for version constraints or conventions that override default patterns.

5. Flag mismatches. If any dependency has known breaking changes between the installed version and what the model's training data likely contains, flag them immediately:

   ⚠ VERSION ALERT: Next.js 16.1 installed — middleware.ts was renamed to proxy.ts in v16. Do NOT use middleware.ts patterns.
   ⚠ VERSION ALERT: React 19 installed — forwardRef is no longer needed, ref is a regular prop.
   ⚠ VERSION ALERT: Drizzle ORM 0.35+ — schema definition API changed significantly from 0.28.
   

6. Spec-anchor check (intra-repo claims). Steps 1–5 verify external dependencies. Step 6 verifies internal ones — concrete repo identifiers carried in from /shape that the PRD will treat as ground truth. The same false-confidence failure mode applies: a model can name a file path, table, function, exported symbol, or test file that doesn't exist, and downstream skills will inherit the phantom anchor as a contract.

   Enumerate the identifiers /shape referenced — file paths, schema/table names, function or type names, test files, named patterns ("the existing X loader," "the canonical Y format"). For each, confirm it resolves in the current repo:

   # File path or directory
   test -e <path> || echo "MISSING: <path>"
   
   # Exported symbol (function, type, class, const)
   rg -n --type ts "export (function|const|class|interface|type) <name>\b"
   
   # Schema table or named structure
   rg -n "<table_name>" drizzle/schema.ts  # or equivalent for the project
   

   Flag every identifier that doesn't resolve, and flag every vague-noun reference that resists grepping ("the chunks table" without a name, "the existing pipeline tests" without a path, "the same loader path the reranker uses" without a function). Vague nouns are the canonical site of spec-rot — they survive /execute Step 0's Consumes-verification gate (which only fires on named symbols) and only surface at /pre-merge. Resolve each by either pinning the exact identifier or rewriting the claim in terms that don't pretend an entity exists.

   Format mismatches as spec alerts so they are visually distinct from version alerts:

   ⚠ SPEC ALERT: PRD references "knowledge_chunks table" — no such table in drizzle/schema.ts. Closest match: knowledgeObjects (chunks generated in-memory by src/scripts/embed-knowledge.ts).
   ⚠ SPEC ALERT: PRD says "mirroring existing pipeline tests" — no test file exists at src/mastra/rag/*.test.ts. Pattern is invented, not extended.
   

   Keep this within Phase 0's 30–60s budget — only check identifiers actually surfaced in /shape, not every name in the repo.

Phase 1: Auto-Calibrate Depth

Based on Phase 0 results, determine the research depth:

TARGETED (2-5 min) — Use when:

• Phase 0 found no version surprises
• The feature extends existing patterns already in the codebase
• No external APIs or unfamiliar libraries involved
• The approach is clear from the shape session

For TARGETED depth: write a short research document (20-50 lines) containing the version check results, any relevant past solutions from docs/solutions/, and a confirmation that the planned approach is valid for the installed versions. Skip the full template sections (Don't Hand-Roll, Options Evaluated, etc.). Move on to Phase 5.

STANDARD (10-20 min) — Use when:

• Integrating with a familiar external service using a new feature
• One dependency has a version mismatch that needs investigation
• A single technical decision needs evaluation (but not a wide-open choice)

DEEP (20-30 min) — Use when:

• Phase 0 found significant version mismatches
• The feature involves an external API or service not yet in the codebase
• Multiple valid technical approaches need evaluation
• An unfamiliar library with complex documentation is involved

Tell the user which depth was selected and why. If they disagree, adjust.

Phase 1.25: Library Callback Contracts

Only runs when the feature implements a library-provided callback surface. Triggers include: agent prepareStep/onStepFinish/before*/after* hooks, Express/Koa/Next middleware and proxy functions, AI SDK tool execute handlers, render props and higher-order components, lifecycle methods, interceptors, or any other shape where the library hands the application a callback to implement.

Phase 0 verifies APIs the feature calls. Most features stop there. When the feature also implements a callback the library defines, TypeScript's excess-property check does not run on returned values typed against local wrapper interfaces — any field the library's signature does not declare is silently discarded at runtime, even when the build passes. Pin the contract here, once, so /write-a-prd cannot codify a non-existent mechanism and /execute cannot wrap the return in a superset type.

For each callback the feature will implement:

1. Open the installed type definition. Grep node_modules/<library>/**/*.d.ts (or the equivalent for the language) for the callback's type name. Note the file path and line number.
2. Record the accepted return shape verbatim. Paste the return-type declaration — every field, optional marker, and union arm. Do not paraphrase.
3. Name the replace-vs-merge semantics. If the return includes a field that hands the library a collection (system messages, headers, tools, middleware stack), confirm in the library source whether the library replaces the existing collection with your return or merges into it. Replace semantics are the highest-risk class: the callback must reconstruct whatever it wants preserved.
4. Flag any wrapper-type plans. If the shape session imagined a mechanism (e.g. "inject via system override") that doesn't appear in the recorded shape, flag it as a design-time veto for the PRD. Do not downgrade to "we'll figure it out in /execute."

This step is gated — it only runs when a callback surface is in play — so Phase 0's 30–60s budget is preserved for features that only call libraries.

Phase 1.5: API-Shaped Work Check

After selecting depth, determine whether the feature needs focused API contract review.

Invoke /api-design-review only if at least one of these is true:

• a new external or partner-facing API is being introduced
• an existing request or response contract is changing
• OAuth, scopes, token model, or webhook verification is involved
• there is real uncertainty about paradigm selection

You may also invoke it for internal APIs with multiple independent consumers when a contract mistake would create broad cleanup cost.

Do not invoke it for ordinary implementation work that merely calls an API without shaping the contract.

If /api-design-review runs, incorporate its verdict into the research document rather than duplicating a second full API review.

Phase 2: Establish Constraints

State the constraints from the shape session. These bound the research and prevent scope drift. Present each constraint one at a time, ask the user to confirm or correct it, then move to the next:

1. What is the feature trying to accomplish? (1-2 sentences from shape)
2. What are the hard constraints? (scale, budget, self-hosted vs managed, latency, auth model, etc.)
3. What decisions are already locked? (framework, language, deployment target, etc.)
4. What version constraints did Phase 0 surface? (list any flags)
5. If timing language exists, what is currently a target, what is currently an estimate, and what is explicitly not yet a commitment?

Do not present these as a batch. State one constraint, get confirmation, then proceed to the next.

Phase 3: Consult Past Solutions

Search docs/solutions/ for relevant past solutions:

grep -rl "relevant-keyword" docs/solutions/ 2>/dev/null

If relevant solutions exist, read them and incorporate their lessons. Note which past pitfalls or patterns apply. This is the compounding benefit — past work makes future research faster and more accurate.

Scope check before transfer. When a loaded solution doesn't state explicit scope or preconditions (the Rule Scope section added by /compound), or when its stated scope doesn't match the current work's structural shape, re-derive the recommendation from first principles rather than transferring it. A rule correct for a 2-step agent loop with a terminal forced tool can invert for a 3-step loop where the same tool is non-terminal — keyword match is not shape match. Verify the shape, don't skip the check.

Staleness check: When reading a docs/solutions/ file, check its volatility YAML field and date. If volatile and older than 90 days, or if the Shelf Life condition appears to have been met, note it as potentially stale before incorporating its lessons. Flag stale docs to the user — they should be updated or deleted.

Phase 4: Research

Research the codebase and relevant technologies. Use sub-agents in parallel when investigating multiple options.

1. Explore relevant existing code — What patterns exist? What integration points are available? What constraints does the current architecture impose?

2. Fetch current documentation — If Context7 MCP is available, use resolve_library and get_library_docs for the specific installed version. If not, use web search targeting official docs for the installed version — never the "latest" version unless it matches what's installed. Also check for llms.txt at the library's documentation site.

3. Verify every API reference — Do NOT trust training data for any framework API call. Every API method, import path, file convention, or function signature in your research output must trace back to either: (a) the installed version's official documentation, (b) a verified web search result, or (c) existing working code in the current codebase. When an API call goes through an SDK that wraps a remote provider (LLM SDKs over model APIs, cloud SDKs over service APIs, payment/auth SDKs over upstream services), the citation under (a) must include the provider's contract docs in addition to the SDK's types — the SDK's type signature is the upper bound of what compiles; the provider's contract is the lower bound of what executes. Concretely: when handing a JSON Schema, Zod schema, tool definition, or structured-output spec across an SDK→provider boundary, cite both the SDK's type for the field (e.g. Output.object({ schema })) and the provider's schema-subset documentation (e.g. Gemini response_schema, Anthropic tool.input_schema, OpenAI Structured Outputs subset). If you cannot verify an API reference, mark it explicitly: "⚠ UNVERIFIED — could not confirm this API exists in [version]."

   Citation required in the written output. For each verified claim, record one of the following in the research document itself — not only in your reasoning: an official docs URL at the verified version, a file path and line number from node_modules/<library>/**/*.d.ts, or a grep result showing the behavior in the project's codebase. "I confirmed this works" without a traceable citation is not verification. The citation is what makes the research auditable by /write-a-prd and /execute. If you cannot produce a citation, mark the claim as Uncertain and elevate it to Phase 2 as a first-priority constraint.

   Calibrate filed-issue evidence. The upstream issue tracker is a first-class source for what is currently broken — not what an API is supposed to be. When Phase 0 surfaced a 📦 YOUNG FEATURE flag, run gh issue list -R <upstream> --search "<feature-keyword>" --state all --limit 20 (or the platform equivalent) before finalizing the recommendation. Audit both open issues and recently-closed ones (closed issues often surface official workarounds or alternative patterns). Fold findings into the Options Evaluated section — a buggy-on-the-current-runtime path is evidence that may flip the recommendation.

   Then, regardless of whether the audit was proactive or a citation arrived from elsewhere: when a citation originates from a filed GitHub issue (or equivalent bug tracker), verify (a) the issue state (open/closed), (b) whether a fix has shipped and in which released version, and (c) whether the current official docs for the installed version contradict the issue. Canonical current docs outrank filed issues when they conflict; closed or resolved issues are not evidence of current state unless you explicitly read the resolution. A filed issue is a hypothesis about a point in time, not a verdict about now — the cheapest counterfactual is to read the current docs page for the feature the issue describes.

4. Evaluate each option against constraints — Score each option against the constraints from Phase 2. Be explicit about which constraints each option satisfies or violates.

For each option investigated, ask:

• What should be proven first?
• What existing patterns should be reused?
• What boundary contracts matter?
• Are there known failure modes that should shape implementation?

Before leaving Phase 4, produce a lightweight estimation-readiness readout:

• What is known now vs still unknown?
• Which unknowns most widen the current estimate range?
• Is the work still best framed by analogy, decomposition, or calibrated actuals?
• Should any timeline statement remain a range or confidence ladder rather than a single-point date?

The goal is not to create a scheduling ceremony. The goal is to prevent research findings from being summarized as false precision.

Discharging code-verifiable Uncertain assumptions. When this phase produces an Uncertain (or Speculative) assumption whose verdict is cheaply settled by 10 lines of throwaway code — does the library actually expose this, does the streaming path emit partial tags, does this format render where we need it — name /prototype FEASIBILITY as the discharge route in the research document and (if appetite allows) run it before handing off to /write-a-prd. The verdict folds back into this artifact by downgrading the assumption tag, so the PRD inherits a verified premise instead of an unverified one. This is the assumption-scale rung of XP's test-first self-similar rhythm: TDD at the minute scale, FEASIBILITY at the assumption scale.

For API-shaped work, explicitly document:

• the API problem statement and impact statement
• the developer consumers affected by the contract
• the operation inventory the API must support
• the paradigm decision and rejected alternatives
• the auth, scope, or webhook verification model if relevant
• whether the proposed change is additive or breaking

Phase 5: Write the Research Document

The research document has two possible storage locations, selected per project. Both carry the same body and the same frontmatter — the difference is where the file lives, not what it contains. Pick the storage mode first, then write.

Phase 5a: Select the Storage Mode

Read the project-level config to determine where this research goes:

# Read the research.storage field from the project-local settings file.
# Default to "archive" if the file or field is missing.
jq -r '.research.storage // "archive"' .claude/settings.json 2>/dev/null || echo archive

The two modes:

• archive (default) — write to a per-user filesystem path outside the repo working tree (~/.claude/research/...). Worktree- and branch-resilient, never committed, never visible to collaborators. Right for solo / single-machine / private projects.
• spike-issue — file the research as a closed-on-creation GitHub issue in the same repo as the PRD, labeled research. Right for public, portfolio, OSS, transparency-themed, multi-contributor, or multi-machine projects where the PRD's audience is GitHub and the research must be openable by that audience without per-user setup. Spike issue visibility follows the repo's visibility — private repos produce private spikes.

Tell the user which mode was selected and why (e.g., "spike-issue mode: .claude/settings.json declares research.storage = spike-issue"). If neither mode is configured, default to archive and proceed.

To opt a project into spike-issue mode, add this to .claude/settings.json (Claude Code ignores fields outside its schema, so this is additive — Skill Kit owns the research namespace):

{
  "research": { "storage": "spike-issue" }
}

Phase 5b: Compose the Frontmatter

Both modes carry identical frontmatter so future readers can judge freshness regardless of where the file lives:

---
date: 2026-04-14
repo: <owner/repo>
feature: <short phrase>
installed_versions_snapshot:
  - <package>@<version>   # list only the key dependencies this research pinned against
  # When the feature imports through a non-default subpath of a package that
  # ships sibling subpaths backed by different runtime drivers (e.g.
  # @libsql/client vs /http vs /sqlite3 vs /web; react-dom vs /client vs /server;
  # drizzle-orm/libsql vs /better-sqlite3; @effect/platform vs -node), record
  # the subpath as a first-class entry. Sibling subpaths are runtime-distinct
  # even when their type signatures match, so a swap between them is a
  # runtime-affecting change disguised as a type-only diff.
  # - <package>/<subpath>@<version>   # constraint: <one-line runtime note, e.g. "HTTP-only — https: and libsql: URLs only">
callback_contracts_snapshot:
  # Only present when Phase 1.25 ran. One entry per library callback this feature implements.
  # - callback: <library>.<callback-name>
  #   file: node_modules/<library>/.../<file>.d.ts
  #   line: <line-number>
  #   accepted_fields: [<field1>, <field2>, ...]
  #   collection_semantics: replace | merge | n/a
---

Phase 5c — Archive mode (default): Write to the Per-User Archive

When the storage mode is archive, write the research document to the per-user archive outside the repo working tree:

~/.claude/research/<repo-slug>/<feature-slug>-<YYYY-MM-DD>.md

Where <repo-slug> is owner-name derived from the git remote (e.g. chrislacey89-skills), <feature-slug> is a short kebab-case phrase naming the feature, and <YYYY-MM-DD> is today's date. Create intermediate directories if they do not exist.

Why outside the repo: the archive is per-user working memory, not team-shared artifact. Branch switches, worktree cleanup, and .gitignore hygiene cannot accidentally destroy it. It is not committed to the project and is never visible to collaborators.

When this mode is selected, skip Phase 5d. The PRD's Research Reference section will emit the archive path.

Phase 5d — Spike-issue mode: File as a Closed GitHub Issue

When the storage mode is spike-issue, file the research as a labeled, closed-on-creation GitHub issue in the same repo as the PRD will live. The PRD's Research Reference section will emit the issue URL (Refs #N) instead of an archive path.

1. Ensure the research label exists (idempotent — first run creates it, subsequent runs are no-ops):

   gh label create research --color BFD4F2 --description "Research spike — closed-on-creation, frontmatter-pinned" 2>/dev/null || true
   

2. Compose the issue body. Take the entire research document (including the YAML frontmatter from Phase 5b) and use it as the issue body verbatim. Frontmatter inside the issue body is rendered as a code block by GitHub but remains machine-readable to downstream skills.

3. Create the issue and capture the URL:

   SPIKE_URL=$(gh issue create \
     --title "Research: <Feature Name> (<YYYY-MM-DD>)" \
     --label research \
     --body-file <path-to-temp-file-with-research-body>)
   

   Use a title format that makes the issue scannable in default issue lists: Research: <feature> (<date>).

4. Close the issue immediately. Closed-on-creation signals "research complete, point-in-time snapshot" — not "archive purged":

   gh issue close "$SPIKE_URL" --comment "Closed on creation — see body for the point-in-time research snapshot. Do not edit; supersede with a new dated spike issue if research changes."
   

5. Cache locally for this session. Also write the same body to the archive path (~/.claude/research/<repo-slug>/<feature-slug>-<YYYY-MM-DD>.md) so the rest of this session and any sibling tools that read the archive can find it. The spike issue is the canonical durable copy; the cache is a session convenience and may be safely overwritten on re-runs.

6. Emit both the spike URL and the cache path when reporting the completed research, and note that the canonical home is the spike issue.

Why a spike issue: lives outside the working tree (no destructive-git risk between write and push), reachable by the PRD's audience (Refs #N syntax), durable across machines (gh issue view N works anywhere), and labeled-and-closed so it stays out of is:open work-to-do views.

Mutability discipline: the issue body is a point-in-time snapshot, not a living document. Do not edit it after creation. If research changes, file a new dated spike issue and update the PRD's Research Reference to point at the new one — the same superseded-by-new-dated-file rule the archive follows.

Include only sections that have real content. A TARGETED research doc might be 20 lines. A DEEP one might be 300. This document is the compressed handoff to /write-a-prd — once written, the downstream skill works from this file, not from the raw research exploration. If this session is running long, suggest starting /write-a-prd in a fresh session using the archive path as input.

Emoji Legend

Use these consistently across all research documents so reviewers can scan quickly:

Inline Documentation Links

Every API recommendation must include a direct link to the official documentation so the developer can click through and audit. Do not just list sources at the bottom — put the link right next to the claim. When web searching or using Context7, capture the exact URL of the documentation page that confirms each API pattern.

Format: [API or pattern name](URL) 🔗

Example: "Use proxy.ts 🔗 instead of middleware.ts."

If you cannot find a direct documentation link for an API reference, mark it: "⚠️ UNVERIFIED — no doc link found for installed version."

Version Change Callouts

When the planned approach relies on behavior that changed between versions, flag it with a 🔄 VERSION CHANGE callout block. Include what the old behavior was, what the new behavior is, which version introduced the change, and a link to the migration guide. This gives the reviewer full context to decide whether they're comfortable with the new pattern.

Format:

🔄 **VERSION CHANGE (v15 → v16):** `middleware.ts` was renamed to `proxy.ts`.
  - **Before (v15):** Root-level `middleware.ts` with `export function middleware()`
  - **After (v16):** Root-level `proxy.ts` with `export function proxy()`
  - **Migration guide:** [nextjs.org/docs/app/guides/upgrading/version-16](URL) 🔗
  - **Why it changed:** CVE-2025-29927 + clarifying that it runs at the network boundary, not as Express-style middleware

Research Document Template

# Research: [Feature Name]

**Date:** YYYY-MM-DD
**Depth:** TARGETED / STANDARD / DEEP
**Domain:** [Primary technology or problem domain]
**Confidence:** HIGH / MEDIUM / LOW
**Constraints from shape:** 🔒 [1-2 sentence summary of locked constraints]

## 📦 Version Check

| Dependency | Installed | Latest | Status |
|-----------|-----------|--------|--------|
| [name] | [version] | [version] | ✅ No issues / ⚠️ Breaking changes |

[If any version alerts were flagged in Phase 0, list them with 🔄 VERSION CHANGE callouts:]

🔄 **VERSION CHANGE (v[old] → v[new]):** [What changed]
  - **Before:** [Old API / pattern / behavior]
  - **After:** [New API / pattern / behavior]
  - **Migration guide:** [link](URL) 🔗
  - **Why it changed:** [Brief rationale]

## 💡 Summary

[2-3 paragraph executive summary. State the primary recommendation and why. Inline-link key documentation references.]

When referencing specific APIs, always link to docs:
- ✅ Use [`functionName()`](URL) 🔗 for [purpose]
- ⚠️ Do NOT use `deprecatedFunction()` — removed in v[X] ([migration guide](URL) 🔗)

## Estimate Readiness

[Include when timeline, budget, or scope confidence is part of the decision. Keep this short.]

- **Target in play:** [Desired budget, event date, or appetite, if any]
- **Current estimate posture:** [Range / confidence ladder / too early to estimate tightly]
- **Commitment posture:** [No commitment yet / conditional after X / safe to commit]
- **Main uncertainty drivers:** [What still widens the range]

## API Review

[Include only when the feature introduces or changes an API contract.]

- **Problem statement:** [What developer problem this API solves]
- **Impact statement:** [What outcome the API enables]
- **Consumers:** [Who integrates with it]
- **Operations:** [The concrete operations the API must support]
- **Compatibility:** Additive / Potentially breaking / Breaking
- **API review verdict:** Proceed / Proceed with constraints / Revise before implementation

## Library Callback Contracts

[Include only when Phase 1.25 ran — this feature implements a library-provided callback.]

For each callback the feature implements:

### `<library>.<callbackName>`

- **Type definition:** `node_modules/<library>/.../<file>.d.ts:<line>`
- **Accepted return shape (verbatim from installed .d.ts):**
  ```ts
  // paste the return-type declaration as written in the installed library

• Collection-field semantics: [replace / merge / n/a — e.g. "systemMessages replaces the full system-message list; the callback must concat the library's passed-in systemMessages arg to preserve advisor persona"]
• Fields commonly imagined but not accepted: [list any fields the shape or PRD draft referenced that don't appear in the accepted shape. If none, omit.]
• Downstream rule: Any boundary-map Produces entry in the PRD that references this callback must cite this snapshot by file:line.

🚫 Don't Hand-Roll

🪤 Common Pitfalls

🪤 Pitfall: [Name]

What goes wrong: [Concrete description] Why it happens: [Root cause] How to avoid: [Specific prevention strategy] (docs 🔗) Warning signs: [How to detect this during implementation]

Options Evaluated

Option A: [Name]

• Fits constraints: 🔒 [Which constraints it satisfies]
• Violates constraints: [Which it doesn't, or "None"]
• Pros: [Specific advantages]
• Cons: [Specific disadvantages]
• Cost/pricing: [If relevant]
• Docs: 🔗 link

Option B: [Name]

[Same format]

💡 Recommended Approach

[Which option and why. Reference the 🔒 constraints that drove the decision. Link to key documentation.]

🔄 Note for reviewer: [If the recommended approach depends on version-specific behavior, call it out here. Example: "This approach uses the v16 proxy.ts convention. If we downgrade to v15 for any reason, this would need to revert to middleware.ts."]

Relevant Existing Code

[Files, patterns, and integration points in the current codebase.]

• path/to/file.ts — [What it does and how it's relevant]

[Diagram suggestion: if research surfaces a multi-module architecture (≥3 modules with non-trivial connections, or a layering or boundary the recommended approach must respect), consider invoking /mermaid to sketch it — a flowchart or C4 diagram — and embed the source in this section alongside the list. The archive entry is consumed by /execute and Ralph AFK loops resuming cold, where a visual reduces the reconstruction cost; the list stays authoritative for grepping by file path. Skip when the architecture is a single module or a flat 2-module hop.]

🔗 Sources

[Every API reference above must trace to one of these. Grouped by confidence.]

✅ Verified (official docs, matches installed version):

• Source title — [What was learned]

✅ Verified (upstream issue tracker):

• [Search URL or issue list URL] — Audited on YYYY-MM-DD. Reviewed open + recently-closed issues. [Summary of what surfaced, or "Zero relevant issues found at on " if the audit was empty.]

⚠️ Partially verified (community source, or version not exactly matched):

• Source title — [What was learned, caveat]

⚠️ Historical / resolved issue (was true when filed, not current state):

• Source title — [What was once true, when it was resolved, and why it is retained here as context rather than evidence]

❓ Unverified (blog post, may be outdated, or no doc link found):

• Source title — [What was learned, risk]

For TARGETED depth, use only the Version Check and Summary sections, plus any relevant pitfalls or past solutions. Skip Options Evaluated, Don't Hand-Roll, etc. But always include inline doc links and version change callouts — those are the highest-value pieces regardless of depth.

If timeline pressure exists, include the short Estimate Readiness section even in TARGETED mode. Keep it to a few lines rather than forcing a full estimation write-up.

### Phase 6: Review with User

Present the research document to the user. Then walk through these review questions one at a time — ask one, wait for the answer, then ask the next:

1. Does the recommended approach match your instinct?
2. Are there options I missed that you'd like investigated?
3. Any version surprises that change your thinking?
4. Where do you want to run `/write-a-prd` — **here** (continue in this session) or in a **fresh session** using the printed handoff line below?

Do not present all four questions at once. Iterate on each answer until the user is satisfied. Question 4 has exactly two answers — do not improvise a third option ("not yet", "later"). If the user is not ready to proceed, return to questions 1–3 to surface what is missing.

**Before handoff, name any remaining feasibility spikes.** If the document still carries `Uncertain` assumptions whose verdicts are cheaply code-verifiable and have not been discharged, list them under a short "Pending feasibility spikes" closing note and name `/prototype` FEASIBILITY as the discharge route. The PRD can still be written with those assumptions tagged — but the named discharge route makes it cheap for the next agent (or this one, in `/execute` Step 0) to settle them before mid-implementation. The cost-of-delay framing is XP's *Defect Cost Increase*: discharging at `/research` time costs minutes; the same assumption discovered at `/execute` time costs hours.

In archive mode, the research file is saved outside the repo — do not commit it to git. In spike-issue mode, the research lives as a closed GitHub issue and a session-local cache copy at the archive path; do not commit the cache copy either.

**At the end of Phase 6, print the runtime handoff line:**

Next session: /write-a-prd Input:

For archive mode, the input is the absolute path under `~/.claude/research/<repo-slug>/`. For spike-issue mode, the input is the closed GitHub issue URL. Print this line whether the user chose to continue here or in a fresh session — the line is the durable handoff either way.

## Handoff

- **Expected input:** clarified problem framing from `/shape`, including choices, assumptions, impositions, and structural signals
- **Produces:** a research artifact carrying verified version and documentation guidance plus an estimate-readiness posture. Storage location depends on the project's `research.storage` setting:
  - **`archive`** (default): file at `~/.claude/research/<repo-slug>/<feature-slug>-<YYYY-MM-DD>.md`
  - **`spike-issue`**: closed GitHub issue labeled `research` in the same repo as the PRD, plus a session-local cache copy at the archive path
- **May invoke:** `/api-design-review` when contract risk is high enough that the API shape needs focused scrutiny before shaping continues
- **Comes next by default:** `/write-a-prd`

## Lifecycle

**The research artifact is a point-in-time snapshot.** It exists to serve the PRD and Ralph loop during active work, and then persists as durable context for future planning in the same area.

- Reference the canonical location in your PRD (archive path or spike issue URL) so `/execute` and Ralph can read it during implementation.
- After the feature ships, the artifact persists automatically — do not delete it. Branch switches, worktree cleanup, and `/compound`'s closeout step cannot touch the archive entry (it lives outside the repo) or the spike issue (it lives in GitHub).
- Stale research actively harms agent performance — a research artifact that recommends Ably v1.2 when v2.0 has breaking changes will steer Ralph in the wrong direction. The frontmatter `date` and `installed_versions_snapshot` let future readers judge whether the snapshot is still trustworthy before they rely on it.
- **Supersession, not editing.** If research changes, write a new dated artifact (new archive file, or new spike issue) and point the PRD at the new one. Do not edit the old artifact in place — the snapshot semantics depend on each artifact being immutable. Spike issues should not gain comments or body edits over time; the closed-on-creation state and the convention against edits are the substitute for the archive's filesystem-imposed immutability.
- This skill does not yet auto-consult prior archive entries or prior spike issues during Phase 0 — that capability is a separate follow-on.