hrt-dark-mode-routing
Single source of truth for where an escalation-worthy finding goes when tempa-spec's dark mode is active (dark.md exists). Each of specs, design, align, and apply points here instead of inlining its own dark-mode carve-out. Says which findings route to hrt-dark-mode-decision-gate, which still go to the user, and which dark.md section records the result.
How do I install this agent skill?
npx skills add https://github.com/hash-id/openspec-schema --skill hrt-dark-mode-routingIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill defines routing rules for an internal development workflow called 'dark mode'. It determines whether findings are handled by an automated decision gate or escalated to a human user, with a specific focus on ensuring that security-critical decisions and permanent changes always require human intervention. No malicious behavior or security vulnerabilities were identified.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
hrt-dark-mode-routing
dark.md existing is the only signal dark mode is active (created by hrt-dark-mode-opt-in). This skill holds the routing rules that were previously copied into every phase instruction. A phase instruction points here with one line; the rules live here.
If dark.md does not exist, this skill does nothing — every finding follows its normal path to the user.
The rule, by calling phase
A finding that would normally escalate to the user is routed as below. Everything that auto-resolves outside dark mode still auto-resolves — dark mode never changes the auto-resolve rule, only where the non-auto-resolved findings go.
| Phase | Routes to hrt-dark-mode-decision-gate | Still goes to the user | dark.md section |
|---|---|---|---|
specs (hrt-adversarial-authoring, artifact: specs) | a single-reviewer UNANCHORED finding | everything hrt-security-pass decided; a one-way door or a change to settled scope (invariant 5) | Specs |
design (hrt-adversarial-authoring, artifact: design) | a single-reviewer UNANCHORED finding | everything hrt-security-pass decided; a one-way door or a change to settled scope (invariant 5) | Design |
align (hrt-align-consistency-review) | every ALIGN DECISION and HIGH finding, except invariant 5 | a one-way door or a change to settled scope (invariant 5) — and VERIFY and WALKTHROUGH are unchanged, and WALKTHROUGH stays the mandatory closing review; the user is also shown dark.md | Align (tag each entry with its cycle number) |
apply Phase 1 (tdd loop) | a seam-mismatch blocker (the gate can weigh an alternative seam against code, design, and specs) | any other Phase 1 blocker | Apply |
apply Phase 2 (hrt-apply-code-review) | every finding that would otherwise surface to the user, except invariants 2 and 5 | a new security suppression flagged by the floor guard (invariant 2); a one-way door or a change to settled scope (invariant 5) | Apply |
Invariants across all phases
- ANCHORED findings, and findings both independent reviewers raised, auto-resolve in the calling skill and never reach this routing at all.
- A finding that is a security requirement or mitigation already decided by
hrt-security-passALWAYS goes to the user, never tohrt-dark-mode-decision-gate. That decision is a hard constraint, not an opposing case to weigh. The same holds for a new security suppression (# nosec,nosemgrep,gitleaks:allow) thathrt-apply-code-review's floor guard flags: it switches off a security check, so it ALWAYS goes to the user. - The
proposalphase's "one change or several?" question (the change-size rule in the interview) ALWAYS goes to the user. The answer depends on the user's capacity and priorities, which no grounding artifact holds. The proposal has no other dark-mode routing: it has no model review, and the interview is the user's own answers. - WALKTHROUGH in
alignstays mandatory and fully manual in every dark-mode variant. - A finding that is a one-way door ALWAYS goes to the user, never to
hrt-dark-mode-decision-gate: a data migration or destructive schema change, deletion of stored data, a change to a published API or event contract, a config or flag removal other code reads, or a step that has to be undone by hand (the same list the review skills use for "needs its own answer"). So does a finding that changes scopeproposal.md(or, inapply, the specs) already settled. A wrong call here cannot be fixed by readingdark.mdafterwards. The calling skill shows these to the user in its usual "needs its own answer" message and waits; everything else keeps routing as in the table. hrt-dark-mode-decision-gateis always a fresh subagent, always writes its reasoning to the matchingdark.mdsection, and never escalates to the user from within itself — it always produces a decision. See that skill for the steel-manning procedure.
Planning-only vs +implementation
dark.md's header names the variant.
- planning-only: routing applies through
align.applyis out of scope — ifdark.md's header is planning-only,applySTOPS and tells the user planning-only dark mode ends at WALKTHROUGH. - +implementation: routing also applies to
applyPhase 1 and Phase 2 as in the table. After Phase 2,dark.mdis presented in full alongside the code diff as a second closing review before the user commits.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/hash-id/openspec-schema/hrt-dark-mode-routing">View hrt-dark-mode-routing on skillZs</a>