[M12] Detect no-yield loops with bounded evidence classification

[Teakowa]

Parent: #99
Discovered by: wrightkit/agent-lab#68

Goal

Add a first-party M12 lint/stability rule that detects loops whose repeating paths do not yield, while distinguishing the static fact "no yield on a repeating path" from stronger runtime-risk claims such as unbounded execution.

Context

Wright v0.1.0 currently has min-wait-loop, which covers loops that yield at the Workshop minimum wait rate, but a loop whose body contains no yield at all can produce zero findings.

This gap was discovered through the first real agent-lab consumer integration rather than synthetic roadmap expansion, so it is valid corpus/consumer evidence for #99.

Repro

Fixture loop-waitless.opy:

globalvar loopCount

rule "waitless loop":
    @Condition getTotalTimeElapsed() > 5

    loopCount = 0
    while loopCount < 10:
        loopCount += 1

wright lint loop-waitless.opy -f json currently returns ok: true with zero findings. The existing Workshop Agent analyzer reports the construct as workshop.performance.waitless-loop.

Required analysis boundary

Do not mechanically copy the existing Workshop Agent severity semantics.

Wright should separate facts that can be proven statically from stronger risk classification:

• Static fact: a repeating control-flow path contains no recognized yield/suspension point.
• Boundedness/iteration evidence: whether the loop is statically bounded, obviously unbounded, or unknown.
• Risk/severity: derived conservatively from the available evidence and documented limitations.

For example, a provably small bounded loop with no yield is not equivalent to while true with no yield, even though both share the no-yield property.

Scope

• Define the Wright-owned rule ID and evidence classification for no-yield loop detection.
• Detect no-yield repeating paths using existing semantic/CFG services where practical.
• Distinguish clearly between:
  • no-yield + provably/obviously unbounded repetition;
  • no-yield + statically bounded repetition;
  • no-yield + boundedness unknown.
• Choose default severity based on evidence rather than treating every no-yield loop as equally blocking.
• Add positive/negative fixtures, including at minimum:
  • an obviously unbounded no-yield loop;
  • a bounded no-yield loop;
  • a loop containing a valid yield that must not fire as no-yield.
• Include the agent-lab repro/evidence in rule documentation or acceptance evidence.
• Preserve structured rule IDs, source spans, severity configuration, and machine-readable output through the M12 lint contract.

Non-goals

• Claiming exact server CPU cost from static analysis.
• Treating all no-yield loops as guaranteed crashes.
• Automatically inserting waits or rewriting control flow.
• Expanding unrelated OPY syntax merely to improve rule coverage.

Acceptance criteria

• wright lint reports a structured no-yield finding for an obviously unbounded no-yield loop.
• A loop with a recognized yield does not produce the no-yield finding.
• A statically bounded no-yield loop is represented with evidence/severity that does not overstate it as equivalent to an unbounded loop.
• Unknown boundedness is represented explicitly rather than guessed.
• Positive/negative regression fixtures cover the rule and its evidence classes.
• At least one agent-lab/real-consumer case is included as linked evidence.
• [M12 acceptance] Verify tooling-first lint and stability platform #100 independently verifies the rule as part of M12 acceptance.

Relationships

• Parent implementation work: [M12] Ship a bounded evidence-backed Workshop stability rule set #99.
• M12 parent: [M12] Tooling-first lint and Workshop stability analysis platform #89.
• Acceptance gate: [M12 acceptance] Verify tooling-first lint and stability platform #100.
• Real-consumer discovery: wrightkit/agent-lab#68.