Skip to content

README.md

Latest commit

 

History

History
41 lines (37 loc) · 4.03 KB

README.md

File metadata and controls

41 lines (37 loc) · 4.03 KB
Raw

no_human docs

Doc What it covers
quickstart.md Start here — from zero to first task in 5 minutes
configuration.md ~/.no_human/config.yaml, every section + default, plus the model picker (GET /api/models, PUT /api/config/models, nh config models)
verification.md The gates that stop a broken change, the bounded loop, the limits
security.md Subscription-auth boundary, the never-merge rule, guards
TELEMETRY.md The canonical, machine-checked contract for every server-side telemetry event no_human can ever send
CAPABILITY_GAP.md The opt-in, default-off capability-gap event contract: what a gap is, the closed vocabulary, the two sinks (issue #20)
INTEGRATIONS_LEGAL.md The recorded legal/compliance position for each third-party integration (Codex auth, sourcing, what is unresolved)
adapters.md Intake (TRACKER/GitHub/GitLab), context, VCS, CI backends
BACKENDS.md The three coding backends (claude, codex, local): switching, credentials, per-mode defaults
eval.md Golden set, replay scoring, scorecard/CI gate, shadow mode
blockers.md Part 22 taxonomy, escalation, wake watcher, nh reply
KNOWN_ISSUES.md Reproduced defects and what a fix must prove; closed entries stay as the record of what the fix proved
INSTALLER.md The packaged apps: how the frozen server + shell are built, verified and installed (all platforms)
WINDOWS.md The Windows app: audit, divergences from macOS, real-Windows test and acceptance runs
LINUX.md The Linux app (.deb + AppImage): build, divergences from macOS, install/verify, acceptance status
desktop.md The Electron shell: how it attaches to a local nh server, or spawns one
DISTRIBUTION.md Where the two shipped artifacts live and how they update
COST_LEVERS.md What measurably moves the token bill, taken from the live DB rather than re-spent
pr-body.md The shape of every pull request no_human opens, and which sections the coder cannot author
UNTRUSTED_PR_REVIEW.md Reviewing an external pull request in a container, on a machine holding an OAuth token

Reference and reports

Point-in-time measurements and decision records. Each is dated, and a report may predate the current code — check its date before treating it as current.

Doc What it covers
NIGHTLY_EVAL.md The unattended overnight regression gate for the whole funnel, ticket in to reviewed PR out
NORTH_STAR_BENCH.md The published benchmark against the operator's real sessions, generated by nh bench publish
REVIEWER_RECALL_METHOD.md How the reviewer catch-rate number is produced: the seeded-defect methodology
NAVIGATION_VALUE.md Whether exploration reads target what a symbol tool could answer: the #114 phase-3 gate, its pre-registered decision rule, and the (non-robust) HALT it returned on a substitute corpus — which does not close the phase (2026-09-10)
ARCH_REVIEW.md Architecture review against the north star, five independent fresh-context agents (2026-07-14)
UI_AUDIT.md Read-only Playwright audit of every flow in the live app (2026-07-13)
BASELINE_M0.md The M0.5 cost baseline every later milestone is measured against (2026-07-10)
design/lease-takeover-vs-orphan-grace.md Decision record for the 600s lease-takeover and row-orphan divergence
design/memory-lifecycle-triage.md Operator runbook for memory retirement: the 45-day auto-archive sweep and flood control