Repository navigation
v0.0.6 — v0.0.5.x cluster (4 quality + audit fixes)
Theme: v0.0.5.x cluster shipped together. Four BACKLOG-tracked follow-ups to v0.0.5, bundled because they're all small fixes / quality-of-life improvements that make every subsequent v0.0.6+ workflow cleaner. Three of the four runs hit the very 600s agent timeout that the fourth fix is making configurable — concrete validation that the friction is real.
Added
--max-agent-timeout-ms <n>CLI flag +RunOptions.maxAgentTimeoutMs?: number(factory-runtime). Default 600_000 (unchanged). Mirrors the v0.0.3--max-total-tokenspattern exactly: field-level addition, string-label CLI validation (NOT a newRuntimeErrorCode). Wide-blast-radius specs can raise the ceiling explicitly.factory.config.json— optional config file at the consumer project root. Specifies defaults forruntime: { maxIterations, maxTotalTokens, maxPromptTokens, noJudge }. Read byfactory-runtime runfrom cwd. Precedence: CLI flag > config file > built-in default.factory initwrites one with documented defaults.@wifo/factory-spec-reviewin scaffoldeddevDependencies(factory-core).factory initnow produces a scaffold wherefactory spec reviewworks on first invocation (previously the dispatch'sfindPackageRootwalk failed because the package wasn't installed)..factory-spec-review-cacheinGITIGNORE_TEMPLATE(factory-core). The reviewer cache no longer shows up ingit statusafter a freshfactory init+factory spec reviewrun.
Fixed
- Harness strips surrounding backticks from
test:paths and patterns (factory-harness). Recurring spec-authoring pitfall —parseTestLinewas passing the literal token (with backticks) tobun test, which never matched any file. Caught twice now (parse-size v1, factory-runtime-v0-0-5). The SPEC_TEMPLATE backtick-guidance BACKLOG entry is now obsolete; spec authors can write either backtick-quoted or bare paths and the harness handles both. factory-implement-report.filesChangedis reliable (factory-runtime). Replaces the simple post-rungit diffwith a pre/post working-tree snapshot (tracked + untracked + content hash). Two failure modes resolved:- False negative on new-file-only runs (plain
git diffdoesn't report untracked files). - False positive on pre-run uncommitted changes (the agent gets attributed for files that were already dirty).
Pre-dirty paths are filtered out — over-attributing the agent is worse than under-attributing.
- False negative on new-file-only runs (plain
Changed
- All six
@wifo/factory-*packages bumped to0.0.6in lockstep (matches the v0.0.5 publish coordination pattern).init-templatesscaffold deps bumped from^0.0.5to^0.0.6. - ROADMAP shift: v0.0.6 was originally themed
/scope-project+ real-product workflow. That theme moves to v0.0.7; v0.0.6 contains the v0.0.5.x cluster shipped here.
Public API surface
Unchanged across every package. Strictly equal to v0.0.5's surface (27 / 18 / ~16 / 19 / 10 / ~7). All v0.0.6 changes are field-level on already-exported types or internal-only helpers.
| Package | Exports |
|---|---|
@wifo/factory-core |
27 |
@wifo/factory-context |
18 |
@wifo/factory-harness |
~16 |
@wifo/factory-runtime |
19 |
@wifo/factory-spec-review |
10 |
@wifo/factory-twin |
~7 |
Test surface growth
@wifo/factory-core: 74 → 77 (+3 init-ergonomics scenarios)@wifo/factory-harness: 56 → 60 (+4 backtick-stripping scenarios)@wifo/factory-runtime: 131 → 138 (+4 filesChanged + 3 agent-timeout scenarios)- Workspace total: 446 → 462
Reconciliations worth knowing
- The cluster shipped as v0.0.6, not v0.0.5.1. Spec ids stay as
factory-{harness,core,runtime}-v0-0-5-{1,2}(they refer to the v0.0.5 follow-up cluster). The published npm version is v0.0.6 because 4-segment versions like0.0.5.1aren't strict SemVer and would be rejected by the registry. - Three of four runs hit the 600s agent timeout while implementing the cluster. The fourth fix (
--max-agent-timeout-ms) is now in for the next run. Net effect: the v0.0.7 work will not hit this — the configurable knob exists. - Lockstep bump even for unchanged packages. spec-review/context/twin didn't change in v0.0.6 but bumped to 0.0.6 anyway. Matches the v0.0.5 publish-coordination pattern; keeps the scaffold's
^0.0.6deps uniformly resolvable.