tvproductions/gzkit / .agents/skills/gz-health-audit
gz-health-audit
VerifiedNamespace router → the four-axis health and integrity audit, read through four dimensions of agent movement (change point, jurisdiction, invariant, escalation). Use when the operator reports that gzkit feels "wobbly", misaligned, or that governance is not holding — or before a release, to check that declared mechanisms still fire. Routes to the concrete axis surfaces in a fixed cheapest-first order; owns the ordering rationale, the reading frame and the budget rule, not the analyses themselves.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 12.4 KB |
| references/constellations.md | markdown | 6.6 KB |
references/constellations.md
View on GitHub# The four constellations
> Lifted byte-identical from `SKILL.md` § The reading frame on 2026-09-19 (GHI #921). The four-dimension table, the caution that binds every cluster, and the pass stay in the skill.
Each dimension is anchored by the sources the conversation named, marked
*(conv.)*. Saltzer and Schroeder are an operator-requested addition on 2026-09-12,
marked *(operator addition)*. Other additions are the agent's proposals on
operator direction 2026-09-12, pending ratification. This is a map of where the
thinking lives, not a completed literature review. Verify a source and its
claimed application before using it to support an audit finding, inside or
outside the repo. Linked sources support the qualified claims below; their
application to gzkit remains an interpretation to test against local evidence.
Each cluster ends by naming the boundary it crosses into
the next, per the conversation: *"Map that terrain rather than just adding
names."*
**Jurisdiction — the modularity cluster** *(conv., elaborated in the
conversation)*. Parnas 1972 *(conv.)*: a module hides a design decision, so the
boundary that matters is a decision, not a file. Dijkstra (separation of
concerns), Wirth (stepwise refinement), Myers and Constantine (cohesion and
coupling), Brooks (conceptual integrity). With Evans *(conv.)* for the bounded
context and Skelton and Pais *(conv.)* for ownership and cognitive load. The
cluster distinguishes design boundaries from authority boundaries. Saltzer and
Schroeder *(operator addition)*, ["The Protection of Information in Computer
Systems" (1975)](https://web.mit.edu/Saltzer/www/publications/protection/Basic.html),
§ I.A.3(f), supply the least-privilege anchor: grant only the authority needed
for the work. For this audit, compare the agent's available mutation authority
with its authorized change surface; a prose allowlist alone does not prove
enforcement. This application complements information hiding without treating
the two as identical. It crosses into contracts: once the
boundary is drawn, what must hold across it is the invariant lens's question.
**Change point — the seams cluster**. Feathers *(conv.)*, *Working Effectively
with Legacy Code* (2004): a [seam](https://www.informit.com/articles/article.aspx?p=359417&seqNum=2)
permits behavior to be altered without editing at that place; characterization
tests and effect sketches help examine existing behavior and impact. Fowler
*(conv.)*, *Refactoring* (1999; 2nd ed. 2018): change that preserves observable
behavior, and the strangler fig for replacing a surface incrementally. Hunt and
Thomas *(conv.)*, tracer bullets: a thin end-to-end slice through the real
system. Beck, *Test-Driven Development: By Example* (2002): the smallest step
that can fail, then pass, then be cleaned. Lehman, ["Programs, Life Cycles, and
Laws of Software Evolution" (1980)](https://users.ece.utexas.edu/~perry/education/SE-Intro/lehman.pdf),
Table I: continuing evolution can increase complexity unless work counteracts
it. This motivates inspecting coupling; it does not establish that age alone
narrows the legitimate seam. The cluster
decides where change is safe. It crosses into impact: a seam whose effect sketch
leaves the authorized surface is the escalation lens's question.
**Invariant — the contracts cluster**. Meyer *(conv.)*, *Object-Oriented
Software Construction* and "Applying 'Design by Contract'" (1992): preconditions,
postconditions and class invariants as obligations, not comments. Brooks
*(conv.)*, conceptual integrity: one coherent design story. Floyd (1967) and
Hoare (1969): assertions and the axiomatic basis, where pre and postconditions
come from. Hoare, ["Proof of Correctness of Data Representations" (1972)](https://link.springer.com/article/10.1007/BF00289507):
representation correctness is a candidate lens for corpus-to-rendition
preservation. That analogy needs an explicit abstract/concrete mapping and
preservation obligations; a renderer is not automatically such a proof.
Gries, *The Science of
Programming* (1981): the loop invariant as the thing that stays true while the
work moves. Liskov and Wing, "A Behavioral Notion of Subtyping" (1994):
contracts survive substitution, or the substitution is wrong. Ford, Parsons and
Kua, *Building Evolutionary Architectures* (2017): the architectural fitness
function: assessment of a named architectural characteristic. A validator
qualifies only when its measured property serves that architectural objective.
Nygard,
"Documenting Architecture Decisions" (2011): the ADR as the artifact whose
agreement with the code is the coherence this lens tests. The cluster decides
what must hold. It crosses into cognition and maintenance: inspect whether the
next session can reconstruct the invariant and its rationale from the artifacts
and handoff, rather than assuming the artifact alone carries understanding.
**Escalation — the impact-and-control cluster**. Arnold and Bohner *(conv.)*,
*Software Change Impact Analysis* (1996): the discipline of estimating what a
change will touch before making it. Yau, Collofello and MacGregor, "Ripple
Effect Analysis of Software Maintenance" (1978): an early treatment of the same
question. Weiser, "Program Slicing" (1984): dependencies relevant to a specified
program behavior. Compare ontology reach cautiously: graph reachability is not
program slicing without an explicit criterion and adequate data/control-flow
semantics. Letovsky and Soloway, "Delocalized
Plans and Program Comprehension" (1986): why impact escapes the seam, because a
plan is spread across places that do not name each other. Rasmussen, "Risk
management in a dynamic society" (1997): systems migrate toward the boundary of
acceptable performance under pressure; test that explanation against observed
pressures and adaptation rather than equating every discrepancy with migration.
Leveson, [*Engineering a Safer World* (2012)](https://mitpress.mit.edu/9780262533690/engineering-a-safer-world/):
safety as a control problem. Use that proposed lens to inspect control actions,
constraints and feedback. The airlock's purpose remains operator canon:
orientation and synthetic memory, movement control, focus and contamination
awareness, and monitoring results/disturbance. The analogy cannot redefine it
as a verification gate. Ohno, *Toyota Production System* (1988): jidoka and the
andon cord, stop the line and make the problem visible, the practitioner root
of stop-and-report. Hutchins, *Cognition in the Wild* (1995): cognition
distributed across people and artifacts, which is what a controlled handoff
carries. The cluster decides when to stop and what the stop must say. It
crosses back into jurisdiction: an impact argument ends by naming the boundary
that should have been drawn.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |