Komodo capability audit — 2026-10-25¶
TL;DR¶
Komodo's foundation is appropriate, but dependency coverage, result-aware orchestration and failure isolation have more value than adding alerts. Live infra-services runs Core/Periphery v2.3.4 and already has Reddit Explorer's input-aware conditional deployment. This initial assessment changed documentation and shared guidance, not production deployment policy. The later deployment uplift records approved implementation, credential preparation and isolated PR packaging; code rollout remains pending.
Evidence and phase status¶
The full audit records exact
production revision, image digests, selected live resource/schedule fields, upstream
sources, limitations and acceptance gates. At the initial capture, the working
branch started at 4e7a8d1; production and the inspected remote-main snapshot were
fec3dbc. Later remote main 7e7f79b6 introduced manual-only dispatch and additional
operational corrections. This journal's diagram and findings describe the earlier
snapshot, not current trigger policy. Do not infer missing production features from
the older branch or overwrite newer contracts when integrating.
| Area | Status |
|---|---|
| PLAN Phase 4 / Komodo ownership | Existing deployment platform retained; no phase completion claim |
| Capability assessment | Repository, production revision and selected live state compared |
| Reddit optimization | Already present in production; reuse rather than rebuild |
| Other stack optimizations | Recommendations only; not implemented or deployed here |
| Alerts | Not a focus; no alert-policy changes |
Documentation changes¶
- Added the capability audit with ranked work, native feature semantics and acceptance gates.
- Corrected the service README, webhook and service-onboarding runbooks: polling is refresh-only, HTTP 200 is not completion, UI write disabling is not API authorization, and the installed version supports Procedure batch tags.
- Linked the current assessment from the root README, architecture index and navigation.
- Added agent guidance for dependency tracking, production comparison and documentation updates; added a supersession notice to the July reliability audit without rewriting historical incident evidence.
- Preserved unrelated working-tree changes and did not modify
PLAN.md, create a branch, commit, merge, or dispatch a deployment.
Execution understanding at the initial capture¶
flowchart TD
A[Resource refresh] --> B[Observe state and pending changes]
C[Eligible main change] --> D[LAN relay requests external Sync]
D --> E[Fixed settling delay - not completion verification]
E --> F[Procedure request]
F --> G[Conditional batch and application stages]
G --> H[Inspect terminal Updates and application acceptance]No autonomous Git-deployment fallback schedule was observed. Three bootstrap Procedures are live outside the seven Git-declared Procedures: database backup, image-update checks and server-key rotation. All observed Stack auto-update flags are false. Five sampled general deployments completed successfully; that does not remove the relay's fixed-delay race or prove every stack's input coverage.
The audit also records an unsuccessful version-probe command: core --version
starts a second Core instead of just printing a version. It exited on the occupied
port; the original containers retained their start times and zero restart counts.
Use existing startup/version metadata next time, not that executable probe.
Validation boundary¶
The original audit's working-tree checks are not integrated PR evidence. Packaging
uses a fresh clone based on 7e7f79b6; no earlier test totals, Markdown counts or
original-checkout duplicate-file blockers are claimed for it. Broad tests and the
site build are deferred until the parent finishes runtime/test integration and
records exact commands/results in the uplift journal.
Safe-pull tests require Bash 4+ and Git. No source check establishes live acceptance.
Recommendations at the audit capture¶
| Priority | Work | Ownership / gate |
|---|---|---|
| High | Extend native tracked-input coverage to source/config-sensitive conditional stacks | Agent implementation with input-mutation tests and reviewed rollout |
| High | Replace fixed sleep with exact Sync/Procedure Update completion | Reviewed API identity and SOPS credential; no runner Docker socket |
| Medium | Reuse Reddit selection pattern for Orders and PR Dashboard | Preserve force retry and explicit build/refresh policy |
| Medium | Isolate host/app failures; keep Firefly refresh ahead of fragile apps | Procedure dependency and failure-injection tests |
| Medium | Reproducible version pins and optional native tag targeting | Verify LXC agents; preserve policy exclusions and disjoint targets |
| Medium | Close validator/hook gaps; document bootstrap schedules | Negative fixtures, ownership and recovery review |
| Later | Evaluate a real missed-trigger reconciler and narrower environment exposure | No frequent full-pipeline schedule before selectivity/result gates |
These are the original recommendations, not the current implementation backlog. The later uplift implements the high/medium source work, retains explicit target selection and chooses manual missed-trigger recovery rather than a recurring full deploy. Keep implemented, deployed and verified status distinct; the uplift's integration validation, bootstrap and live acceptance remain open.