Known Defect Disposition
Severity controls response urgency; it does not authorize neglect.
P0/P1 triggers immediate review and remediation. That part of every triage scheme works fine and needs no new skill. The failure mode this skill exists for lives one tier down: P2, P3, and P4 getting treated as lower priority in a way that quietly becomes not anyone's problem. They are not. They are known defects requiring disposition — the same obligation debt-closure- discipline states for any known issue, applied specifically to the moment a severity tier gets assigned and a decision gets made about what happens next.
A known defect cannot disappear into prioritization. It must be repaired, explicitly accepted, or explicitly deferred with evidence — and deferral requires first testing whether interactions can amplify its severity.
That second clause is the part most triage processes skip, and it is the part this skill is actually about. Initial severity is almost always scored against the finding in isolation — this input, this function, this code path. The reachable consequence is frequently not the isolated finding; it is what the finding becomes once it is placed next to who can trigger it, what else it touches, and what is reachable from the state it leaves behind. Scoring the isolated finding and stopping there is not triage — it is describing the symptom and calling it the diagnosis.
Composes with the library:
- debt-closure-discipline — defines the three terminal states (resolved / explicitly rejected / documented-as-open) every known issue must reach. This skill supplies the mandatory gate that must run before a defect is allowed into the "documented as open" state, and the structured record that state must contain when the defect is a security- or correctness- relevant finding rather than generic debt. Don't restate that skill's three-state mechanism here — compose to it.
- assume-breach-modeling — the escalation questions below ("who controls it", "what's reachable from the stuck state") are that skill's chaining method, run once per deferred finding instead of once per system.
- exploitability-triage — tells you whether an escalation path is
theoretically describable or actually reachable in this deployment; an
escalation hypothesis with no reachability behind it is a
PLAUSIBLE, not aCODE FACT— label it that way. - claim-provenance-discipline — every field in the escalation analysis (Part 3) carries its own epistemic level; don't let a guess about reachable consequence read as a confirmed one.
- decision-record-discipline — a deferral is a decision. It gets that skill's revisit-trigger discipline, not a bare "won't fix."
- beyond-the-fix / red-team-auditing — supply the adversarial posture this skill's escalation gate asks you to apply to your own findings before someone else applies it for you.
- destination-driven-construction — "insufficient time" stops work at the last coherent level; it does not authorize pretending the next level's known gaps don't exist. Same principle, applied here to defects instead of build phases.
Part 1 — What severity tiers actually mean
A tier says when something gets worked and in what order. It does not say whether the lowest tier may sit forever as a semantically-closed item just because someone wrote it down with a "P3" next to it. Restated plainly, because this is the entire thesis:
- P0/P1 → immediate review and remediation. Not in dispute.
- P2/P3/P4 → scheduled later, worked in tier order. Still known defects requiring disposition. Not a quieter queue that disposal obligations don't apply to.
If "not blocking current work" is the only thing written next to a tier, nothing has actually been decided — see Part 4.
Part 2 — The escalation gate: mandatory before any deferral
Before a finding is allowed to leave triage as anything other than "fixed now," it must pass through this gate. Skipping the gate is the single most common way a real P1 ships labeled P3.
Finding
│
▼
Initial severity P0–Pn (scored against the finding in isolation)
│
▼
Can interaction, composition, adversarial control,
deployment context, or another known defect raise severity?
│
├── YES → investigate the escalation path before deferral.
│ Re-score. If the escalated path reaches P0/P1
│ territory, it IS P0/P1 — relabel, don't footnote.
│
└── NO → retain current classification, but retain it
WITH the evidence that was checked, not by default.
│
▼
Fix now?
├─ YES → fix → regression evidence → close (debt-closure-discipline
│ state 1, and beyond-the-fix scrutiny before it counts as closed)
└─ NO → explicit deferral decision (Part 4) → PENDING_BUGS.md
The escalation analysis is not optional ceremony for findings that "feel" serious. It is mandatory for every P2–P4 before deferral, because the entire reason this skill exists is that "feels minor in isolation" is exactly the judgment the gate is built to distrust.
The six questions
Run these against the finding, not against your intuition about the finding. Answer each one with evidence (a trace, a code reference, a reproduction), not with a plausibility argument alone — an unevidenced "probably not" is how the gate gets rubber-stamped instead of run.
- Who controls the value or trigger condition? An attacker-reachable input and a value only your own deploy pipeline ever sets are not the same finding, even if the code path is identical.
- Can it be induced deliberately, on demand, not just hit by accident? A bug that requires a one-in-a-million race is a different risk than one a counterparty can trigger at will by sending one crafted message.
- What state is left behind once it fires — and is that state stuck? "Reverts cleanly" and "leaves the system in a state nothing can exit" are different findings wearing the same stack trace.
- Is value — funds, credentials, trust, authority — reachable from that stuck state? This is where "overflow → revert" becomes "overflow → activation permanently blocked → counterparty appropriates the position without posting collateral." The syntactic bug and the reachable consequence are not the same sentence.
- Does an alternate path exist that avoids the stuck state, or is this the only way through? A recoverable dead end is lower severity than an irrecoverable one even when the trigger condition is identical.
- Does the party who can trigger this gain an asymmetric economic or security benefit from doing so? A bug that only hurts its trigger is a different risk posture than one that pays the trigger to pull it.
Worked shape of the pattern (generalized from the kind of finding that motivates this skill — a numeric overflow superficially read as "DoS, moderate"): asking who controls the overflowing value, whether they can force it deliberately, and what is reachable from the resulting revert can turn "integer overflow causes a revert" into "a counterparty can force the activation step to fail permanently, and the frozen state lets them acquire the position at the stale price with no collateral posted." The first sentence describes a crash. The second describes a theft. Both describe the same line of code. Only the gate tells you which one you are actually holding.
If any question's answer pushes the reachable consequence into P0/P1 territory, that is the severity — not a caveat attached to the lower number. A footnoted P0 is a mislabeled P0.
Part 3 — "Not blocking current work" is not a disposition
This phrase, by itself, describes scheduling. It says nothing about the defect's disposition under Part 4's three terminal states. It is routinely used as if it were a decision, and it is not one — it has no reason, no accepted impact, no revisit trigger, and no record of whether the escalation gate ever ran.
MUST NOT treat "not blocking" as equivalent to any of: reject-with-reason, documented-as-open, or fixed. It is a true statement about sequencing that carries zero disposition content on its own, and it must never be the only thing recorded about a deferred finding.
Part 4 — The structured deferral record
A defect that is not fixed now is still required to leave triage with a
record carrying: initial severity, the escalation analysis actually run (not
"assumed clean"), the disposition, and the conditions that would reopen it.
This is decision-record-discipline's revisit-trigger discipline, applied
to a defect instead of a design choice. Store it in a tracked
PENDING_BUGS.md (or your project's equivalent known-issues ledger) — not a
comment, not a chat message, not a verbal agreement.
id: BUG-NNNN
title: <one line, states the defect, not the symptom>
severity_initial: P3 # scored against the finding in isolation
severity_reviewed: P3 # after the escalation gate; may differ from initial
escalation_analysis:
adversarial_control: <who can trigger this, and how>
composition: <what else this interacts with>
affected_invariants: <what property breaks if this fires>
reachable_consequence: <the worst reachable state, traced, not guessed>
p0_p1_path_found: false # true forces severity_reviewed up — see Part 2
evidence: <trace / repro / code reference the above rests on>
disposition: DEFERRED # one of: FIXED | REJECTED | DEFERRED
reason: <why not fixed now — a real constraint, not "didn't get to it">
why_not_fixed_now: <time, dependency, scope — named specifically>
known_impact: <what is actually exposed while this stays open>
accepted_exposure: <what you are explicitly choosing to tolerate, and for whom>
revisit_trigger: <observable condition that reopens this — a date,
a traffic threshold, a dependency shipping, not "later">
owner: <who is accountable for the revisit, not just the fix>
A record with escalation_analysis filled in but p0_p1_path_found never
actually evaluated — left as a default false nobody checked — is a record
that looks complete and is not. The field exists to be answered, not to be
inherited as a template default.
Part 5 — Insufficient time is a reason not to fix; it is never a reason to erase
Running out of time, a dependency not yet available, or a lower-ranked priority queue are all legitimate reasons a defect stays open. None of them are a reason for the defect to stop existing in the project's record.
This is the same discipline destination-driven-construction applies to
build phases — when time runs out, you stop at the last coherent level, you
do not pretend the next level's requirements were never real. Applied here:
when time runs out on triage, the unresolved findings go into
PENDING_BUGS.md with the record in Part 4. They do not go nowhere.
"Deferred" must remain visibly, trackably open — exactly
debt-closure-discipline's state 3, never read as a quieter state 4 that
exits tracking.
Failure modes / anti-patterns
- Isolated scoring without the gate. Severity assigned against the finding alone, with no attempt to ask whether interaction or adversarial control changes the picture — the single most common way a P1 ships wearing a P3 label.
- "Not blocking" used as a disposition. See Part 3 — it is a scheduling statement, not a decision, and recording only this is recording nothing.
- Escalation field present but unanswered. A
PENDING_BUGS.mdentry withescalation_analysiskeys filled with placeholders or copied boilerplate instead of an actual trace of reachable consequence. p0_p1_path_found: truefootnoted instead of acted on. If the gate finds an escalation path into P0/P1 territory, the severity is P0/P1 — treating the finding as still-deferrable at the old tier because the path is "theoretical" contradictsp0_p1_path_foundbeing true in the first place; downgrade the field, not the severity.- Deferral with no revisit trigger, or one that is not observable ("revisit later" instead of "revisit if monthly volume exceeds X" or "revisit when dependency Y ships").
- The record living somewhere untracked — a Slack thread, a verbal agreement, a comment in a PR that gets squashed away — instead of in a file that survives the session that created it.
- Treating time pressure as grounds to drop the finding from the record, rather than grounds to leave it open and say so.
Verifiable checks
- Every P2–P4 finding has a documented escalation analysis (Part 2's six questions, answered with evidence) before it is deferred — not scored in isolation and left there.
- No finding is marked deferred on the strength of "not blocking current work" alone; a reason, known impact, accepted exposure, and revisit trigger are all present.
- Any escalation analysis that found a path to P0/P1 consequence has had
its
severity_reviewedraised accordingly — not footnoted at the original tier. - Every deferred finding exists as a structured record (Part 4) in a tracked file, not only in conversation, a comment, or memory.
- Every deferral record has an observable revisit trigger, not "later" or "someday."
- A defect dropped due to insufficient time still appears, open, in the project's known-issues ledger — it was never silently removed from the record because it didn't get fixed.
- A "fixed" disposition has been checked against
beyond-the-fixbefore being marked resolved, consistent withdebt-closure- discipline's state 1.
How to respond when this skill is active
- When a finding is tiered P2–P4, run the six escalation questions before accepting the tier — out loud, with evidence, not as an assumed "probably fine."
- If any question surfaces an adversarially-reachable path to a worse consequence, say so plainly and re-score — do not present the escalated risk as a footnote under the original number.
- Refuse to accept "not blocking current work" as a complete disposition; ask for the reason, the known impact, the accepted exposure, and the revisit trigger, and write them down.
- Push every deferred finding into a structured
PENDING_BUGS.md-style record (Part 4) rather than leaving it in chat history or a verbal agreement. - When time runs out before every finding has a disposition, say explicitly which findings remain undecided and file them as open — never let "we ran out of time" read as "these are resolved" or let them quietly disappear from the list.