Feasibility Spike

Feasibility Spike

> Part of the discovery-phase skill pack · validation group · reads discovery-context.md (run profile-builder first if missing).

Kill the biggest technical risks cheaply. Throwaway by contract.

Step 1 — Read context

Read discovery-context.md (section 3. Engagement → Runner role) and risk-assumption-map.md (anchors the spike to a specific assumption).

If discovery-context.md is missing, ask the BA inline: "who's running the spike — solo BA / BA + engineer / engineering lead?" — tag the output [ASSUMED ENGINEERING]. If risk-assumption-map.md is missing, ask: "name 1–3 feasibility assumptions to test in one sentence each" or proceed as open exploration tagged [EXPLORATION-NOT-SPIKE]. Never block; recommend profile-builder / risk-assumption-mapping for high-stakes work.

Engineering capacity reality-check (warn + confirm, do not halt): if runner is solo BA with no engineering support, warn that the spike will produce a spike-shaped artifact with weak engineering signal. Offer: loop in an engineer, defer the spike to delivery, skip and accept the risk in the proposal, or proceed solo with a ⚠ NO-ENGINEER-OVERRIDE banner at the top of tech-spike-report.md and findings tagged [SPECULATIVE].

Step 2 — State spike goal

One sentence. Tie to a specific assumption from risk-map.

> Example: "Prove we can extract structured events from free-text emails with ≥80% precision using a single Claude API call."

Step 3 — Time box

| Tier | When | |---|---| | Short — 4 hours | Ruling out a "is this even possible?" question | | Standard — 1-2 days | Validating a specific approach against representative inputs | | Escalated — up to 1 week | Genuinely unknown territory; requires explicit BA + engineering lead sign-off | | >1 week | This is no longer a spike. Re-frame as prototype-plan. |

Stop at the time box. Overrunning spikes are how discovery bleeds into delivery.

Step 4 — Narrowest test

• Use production-representative inputs, not cherry-picked
• Throwaway branch / repo named spike/ — never merge to main
• Prefer manual scripting over frameworks (frameworks hide what you're testing)
• Logging is essential; UI is irrelevant

Step 5 — Build minimum, run, measure

Capture:

• What worked, with evidence (screenshots, logs, numbers)
• What didn't, with same
• Surprises — especially newly-discovered constraints
• Rough performance, latency, cost per operation

Step 6 — Feasibility call

Three outcomes only:

• Green — proceed to PoC / MVP as scoped
• Yellow — feasible with caveats; caveats become PoC scope constraints
• Red — blocker. Escalate before building anything. Options:
• Re-scope (loop feature-scoping)
• Different solution (loop opportunity-mapping)
• KILL or PIVOT in go-nogo-memo

In outsourcing, a Red call before contract = save the engagement (don't sell what can't be built). After contract = harder; either renegotiate scope or eat the loss.

Step 7 — Archive + tear down

Commit spike code under ./discovery/spike//. Do NOT merge to product code. Document tear-down status — was the spike code disposed of, kept as reference, or repurposed?

Output

• ./discovery/tech-spike-report.md per ./template.md
• ./discovery/spike// — disposable code

Append to _log.md: [feasibility-spike | YYYY-MM-DD] goal: ; box: ; result: ; constraints: .

Anti-patterns

• Spike without an assumption. That's exploration. Useful sometimes, but don't mislabel.
• Merging spike code. Spike code skips error handling, tests, polish. Shipping = silent tech debt.
• Overrun. "Just one more hour" violates the rule. Stop, report yellow, decide.
• Hiding red. A failed spike is a saved project. Surface it; celebrate it.
• Spike with no time box. That's R&D. Different cost structure.
• Skipping prerequisite check (engineering capacity in context). BA running this solo will produce a spike-shaped artifact with zero engineering signal.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: stanislavnianko
• Source: stanislavnianko/product-discovery-claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.