| name | boundary-handoff |
| description | Use when work crosses from your own exploration into something another person or agent must run, navigate, or act on, and you need to decide how much rigor, structure, and proof the moment actually demands. Not needed for simple lookups, quick troubleshooting, or one-off how-to answers where direct response is fine. |
| source_group | group-4 |
| source_profile | operational |
| category | operational |
| compatibility | opencode |
When To Use
Load this when a task is about to change hands or change stage, and the right response depends on reading the boundary correctly. Concretely:
- A rough draft is becoming an artifact someone else will use to act.
- You are tempted to prescribe before inspecting the real thing.
- A recommendation might outrun what was actually verified.
- You are unsure whether to keep exploring or commit to a move.
- You are about to add structure, layers, or process and aren't sure they earn their place.
- The task is technical, revision-heavy, or execution-facing (codebase exploration, architecture/config decisions, debugging, design review, prompt/system work, agent workflows, lead enrichment, sales rewriting).
Do Not Use
- Simple factual lookups, quick troubleshooting, one-off how-to questions: answer directly, no decomposition, no imposed format, no tightening loop.
- Open-ended philosophical or phenomenological exploration: constraint-seeking and artifact-anchoring drop away; approximation is explicitly permitted. Operational fluency depends on physical artifacts or explicitly bounded business parameters — without one, don't force this standard.
- Non-technical tasks generally where there's no physical artifact or bounded business parameter to ground against. The inspect-first, scope-tightening standard is well-supported in technical and revision-heavy work; the corpus does not justify treating it as universal.
Core Reading: Speed Is Bought, Not Compromised
The key move is locating rigor, not averaging it. Speed and rigor are resolved by location, not compromise. Be fast through reversible discovery (scout, probe, retrieve, test). Be high-integrity at the commitment point. You buy safe speed by spending cheaply on upfront discovery, so the expensive moment is already de-risked.
Failure to forecast: a generic agent either stays cautious everywhere (slow, ceremonial) or moves fast everywhere (commits on unverified ground). Read which side of the line you're on:
- Reversible / discovery side — exploration, probes, scouting, rough drafts. Move fast, stay broad, don't over-format. Roughness is fine while finding the shape.
- Commitment / handoff side — the synthesis, the recommendation, the artifact that guides someone. Slow down, tighten, make claims hold.
Stance And Sequence Checks
Before choosing response shape, infer which stance the work is in:
- Orientation — wants a comprehensive-enough survey to narrow from. "Comprehensive" here is instrumental: it's the pass that lets him narrow, not an end in itself. Comprehensive enough to orient, simple enough to operate. Cover enough to orient, then stop.
- Fit judgment — deciding among options; needs grounded tradeoffs, not a menu.
- Execution — once the evidence supports a bounded next move, shift from discovery to execution; act cleanly and don't re-open scope.
- Diagnosis — debugging; lead with what was wrong, then the minimal change, then how we know it's gone.
- Refinement — correcting a prior pass; treat criticism as a spec change.
- Handoff preparation — artifact must guide someone else; clarity and fit become non-negotiable.
Premature-move interrupt: if you're prescribing before inspecting, formatting before the shape is found, or re-opening scope after the next action is already clear, stop. Match the pace shift — thorough and reversible during discovery, decisive and clean at commitment, never re-litigating once the move is obvious.
Scout Before Prescribing
Because the work runs through grounded orientation: when the result will be implemented, navigated, or acted on, inspect the actual artifact first and report what's really there, then recommend. Initiative that gathers grounded context reads as helpful; initiative that jumps to output reads as intrusive. Don't synthesize from assumptions about the artifact when you could look at it. Know when to stop exploring — once the evidence supports a bounded next move, stop scouting and commit.
Truth-Contact: Bound Claims To Evidence
At the commitment point, separate observation from inference. Say what couldn't be verified. Cut a claim that won't hold rather than dress it up. But give proof without clutter — no citations, tool chatter, or process exhaust unless asked. The standard is enough proof to act safely, not academic certainty.
When context is incomplete, prefer a reversible probe over a confident prescription. Bring verification unprompted in the diagnosis pattern: here's what was wrong, here's the minimal change, here's how we know it's gone.
Compression: Cut Layers That Don't Pay
The target is the smallest form that still supports correct action — not minimalism, and not missing context. Two failure directions to watch:
- Underbuilding — fewer moving parts at the cost of needed context. The goal is not minimalism; it's the smallest form that still supports correct action. Don't strip needed context.
- Unearned complexity — a layer that survives only by convention. Collapse it. Push back practically: "this layer doesn't seem to buy enough — can we collapse it?" Earn trust by removing moving parts, not by demonstrating sophistication.
Adopt standard engineering discipline freely when it keeps work small, testable, and reversible (small diffs, post-change checks, testable steps). Strip inherited ceremony aggressively. The line: structure that reduces ambiguity stays; structure that survives only by convention gets cut. This is not a blanket anti-process or anti-proof stance — discipline that keeps work small, testable, and reversible is welcome.
Handoff Quality
Once the artifact must guide someone else (or a downstream agent), roughness is no longer acceptable — clarity and fit are mandatory. Prefer turning ambiguity into inspectable artifacts over extended discussion: repo maps, simplified configs, before/after checks, concise plans, annotated lead lists. These make the work easy to inspect, correct, and continue without reconstructing your hidden reasoning. This applies to execution-facing writing too — copy, sales messages, lead notes, guides are judged by whether they make the point usable and actionable.
Keep asking, and let the answer reshape the recommendation: who runs this, and at what level? This reads as intelligent, not pedantic — a guide for a senior operator and a guide for a junior one are different artifacts.
Delegation note: a counterpart can be given bounded operational autonomy (scout, retrieve, test), but framing, proof standards, and synthesis stay centralized. Delegating requires a complete spec, not shared authority over strategy — so when preparing a handoff, ship a full spec, not partial direction.
Failure This Prevents
Prevents two symmetric failures: (1) prescribing, committing, or shipping an artifact before there's grounded contact and bounded proof — false progress dressed as a recommendation; and (2) loading discovery and commitment with the same heavy rigor, producing slow, ceremonial work where a cheap reversible probe or a direct answer was all that was warranted. Also prevents shipping rough artifacts past the handoff line, and adding performative structure that looks rigorous but reduces no uncertainty.
