القائمة
Use when 3 consecutive core experiments or optimizations fail to meet expectations — forces honest status assessment and escalates the pivot/downgrade/kill decision to the user
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use when designing the evaluation protocol for a method development project — locks primary metrics, datasets, seeds, statistical tests, and baseline list into an immutable contract
Use when G2 (plan freeze) and G3 (execution readiness) gates have passed — handles all implementation, baseline reproduction, experimentation, and iteration for both Type M and Type D projects; this is Phase 4
Use when research direction is validated (G1 passed) and the project needs method design, analysis framework, evaluation protocol, or story line — this is Phase 3 and must complete before any code or experiments
Reusable protocol for multi-agent discussions that iterate until convergence. Used by results-integration (Phase 5) and paper-writing (Phase 6) to ensure discussions actually resolve issues rather than just noting them.
Use when and ONLY when the user explicitly requests paper writing and G4 (write-ready) gate has passed — handles LaTeX structure, senior-level writing, reference verification, and iterative refinement
Use when Phase 1 (research-direction-exploration) is complete and G1 gate has passed — subjects the research problem to adversarial defense-style questioning, classifies project intent, re-evaluates venue target, and produces a final feasibility ruling before any method design begins
| name | pivot-or-kill |
| description | Use when 3 consecutive core experiments or optimizations fail to meet expectations — forces honest status assessment and escalates the pivot/downgrade/kill decision to the user |
Repeated failure is signal, not noise. When 3 consecutive core experiments fail expectations, stop iterating and escalate. The agent does not get to decide whether to continue — the user does.
Core principle: Honesty over optimism. Escalate over iterate.
Violating the letter of this rule is violating the spirit of this rule.
AFTER 3 CONSECUTIVE FAILURES, STOP EXPERIMENTING. ESCALATE TO USER.
No fourth attempt. No "quick fix." Escalate.
digraph pivot_or_kill {
rankdir=TB;
detect [label="3 consecutive failures\ndetected" shape=doublecircle];
summarize [label="Summarize status\n(what was tried, what failed,\ntrend direction)" shape=box];
analyze [label="Analyze root cause" shape=box];
present [label="Present 3 options\n(Pivot / Downgrade / Kill)" shape=box];
decide [label="User chooses" shape=diamond];
pivot [label="Pivot\nChange method direction" shape=box style=filled fillcolor="#d4edda"];
downgrade [label="Downgrade\nLower venue target" shape=box style=filled fillcolor="#fff3cd"];
kill [label="Kill\nTerminate and archive" shape=box style=filled fillcolor="#f8d7da"];
detect -> summarize;
summarize -> analyze;
analyze -> present;
present -> decide;
decide -> pivot [label="pivot"];
decide -> downgrade [label="downgrade"];
decide -> kill [label="kill"];
}
State plainly:
Do NOT soften language. "Results declined across all three attempts" is correct. "Results showed some variability" is evasion.
Classify the failure into one of these categories:
| Category | Description | Signal |
|---|---|---|
| Method flaw | Core assumption is wrong | All variants fail similarly |
| Data issue | Insufficient, noisy, or wrong distribution | Results vary wildly or ceiling is low |
| Implementation bug | Code-level error | Should be caught by debugging, not this skill |
| Evaluation too strict | Unrealistic expectations | Baselines also fail to meet the bar |
Be specific. "Something is off" is not a root cause.
Present all three. No filtering, no pre-selecting.
a) Pivot — Change method direction.
b) Downgrade — Lower venue target.
c) Kill — Terminate the project.
THE AGENT CANNOT MAKE THIS DECISION. WAIT FOR USER.
Present the options. Stop. Do not nudge. Do not recommend. The user decides.
| Excuse | Reality |
|---|---|
| "One more try might work" | You said that twice already. Escalate. |
| "I think I see the issue now" | You thought that before too. Escalate. |
| "Killing the project is too drastic" | Killing saves resources for better projects. It's responsible, not drastic. |
| "Let me just tune hyperparameters" | Tuning is not innovation. If 3 approaches failed, the problem is deeper. |
| "The baseline is just very strong" | That's useful information. Maybe this gap isn't solvable with your approach. |
3 failures → stop → summarize → escalate → user decides
Honesty is not pessimism. Escalation is not failure. Continuing blindly is.