refactor: hard-fork GSD -> MSD (Make Software Done)
Mechanical rename produced by scripts/msd-rename.cjs: gsd/Gsd/GSD -> msd/Msd/MSD across contents and paths, upstream package/repo coordinates -> @golem15/msd-core and golem15com/msd-core. Deep links into upstream history, sibling upstream packages, the GSD-2 import feature, CHANGELOG.md and .changeset/ are kept as-is. Hand edits on top: MSD block-letter banner and logos, LICENSE copyright line, package/plugin identity, regenerated lockfile, install-tree fixtures, derived registries and benchmark baseline; migration checksum baseline re-locked (MSD keeps its own install state, so no install had applied the old sums); sort-order and regex-escaped expectations in tests adjusted.
This commit is contained in:
@@ -8,7 +8,7 @@
|
||||
|
||||
## Context
|
||||
|
||||
131 of roughly 705 test files already drive the real `gsd-tools` binary through `runGsdTools` in `tests/helpers.cjs` — CLI-altitude testing is solved and well trodden. What no existing test does is carry one project's accumulating state across all five loop steps in a single run. Every loop test today asserts a step in isolation, against state fabricated for that step alone. `CONTEXT.md`'s `RULESET.TESTS.feedback-loop-convergence` already names the alternative — assert convergence across an accumulating sequence — and one instance of it exists for estimation (`tests/estimate-loop-convergence.test.cjs`). This ADR extends that pattern to the loop itself.
|
||||
131 of roughly 705 test files already drive the real `msd-tools` binary through `runMsdTools` in `tests/helpers.cjs` — CLI-altitude testing is solved and well trodden. What no existing test does is carry one project's accumulating state across all five loop steps in a single run. Every loop test today asserts a step in isolation, against state fabricated for that step alone. `CONTEXT.md`'s `RULESET.TESTS.feedback-loop-convergence` already names the alternative — assert convergence across an accumulating sequence — and one instance of it exists for estimation (`tests/estimate-loop-convergence.test.cjs`). This ADR extends that pattern to the loop itself.
|
||||
|
||||
## Decision
|
||||
|
||||
@@ -18,11 +18,11 @@ Build a walk driver, `tests/qa/loop-walk.cjs`, that runs the five loop steps in
|
||||
|
||||
### 2. Layer over `tests/helpers.cjs`; do not build a second substrate
|
||||
|
||||
`runGsdTools` already invokes the real subprocess with a 60s timeout and env injection, and 131 files depend on it. The walk driver is built strictly on top of it. The rejected alternative — a standalone harness re-implementing subprocess invocation — was considered and rejected as duplication (Gall's Law: a working complex system evolves from a working simple one).
|
||||
`runMsdTools` already invokes the real subprocess with a 60s timeout and env injection, and 131 files depend on it. The walk driver is built strictly on top of it. The rejected alternative — a standalone harness re-implementing subprocess invocation — was considered and rejected as duplication (Gall's Law: a working complex system evolves from a working simple one).
|
||||
|
||||
### 3. The stub agent seam, and its limit stated plainly
|
||||
|
||||
`gsd-tools init new-project` (and the other loop init commands) write nothing — they are read-only context projections; the agent is the one that authors artifacts (STATE.md, phase files, etc.). A scripted stub writer standing in for the agent is therefore faithful to the real division of labor, and the entire engine — routing, state transitions, gates — runs for real behind it.
|
||||
`msd-tools init new-project` (and the other loop init commands) write nothing — they are read-only context projections; the agent is the one that authors artifacts (STATE.md, phase files, etc.). A scripted stub writer standing in for the agent is therefore faithful to the real division of labor, and the entire engine — routing, state transitions, gates — runs for real behind it.
|
||||
|
||||
The limit is stated plainly and not glossed over: this walk proves the engine's state machine is coherent across all five steps. It proves nothing about whether a real LLM agent emits artifacts of the shape the walk's fixtures assume.
|
||||
|
||||
@@ -66,7 +66,7 @@ Scenario artifacts live at `tests/qa/fixtures/`, template-derived, carrying an e
|
||||
Running the walk driver against the real engine — not a mock — produced 0 violations and three smell classes on this first pass:
|
||||
|
||||
- `value-hygiene` — `init` returns `agents_dir` pointing outside the project tree.
|
||||
- `untyped-success` — `smart-entry` emits prose unconditionally (`gsd-core/bin/lib/smart-entry.cjs:577-587`), so the routing oracle cannot assert on it until a `--json` mode exists.
|
||||
- `untyped-success` — `smart-entry` emits prose unconditionally (`msd-core/bin/lib/smart-entry.cjs:577-587`), so the routing oracle cannot assert on it until a `--json` mode exists.
|
||||
- `soft-error-exit-zero` — `state-snapshot` reports a missing STATE.md through a payload key with exit 0.
|
||||
|
||||
These are observations for a maintainer to weigh, not defects this PR fixes.
|
||||
@@ -79,14 +79,14 @@ These are observations for a maintainer to weigh, not defects this PR fixes.
|
||||
- **Known limits, stated rather than hidden.**
|
||||
- The stub agent is not an LLM. The walk validates the engine's state machine; it says nothing about whether a real agent produces artifacts of the assumed shape.
|
||||
- `ship` has no runnable predicate evaluator (`tests/loop-hooks-ship-pre-e2e.test.cjs`: "enforcement is ship.md prose only"), so the `ship` step is asserted at contract-shape level only, not behaviorally.
|
||||
- `smart-entry` emits prose unconditionally on success (`gsd-core/bin/lib/smart-entry.cjs:577-587`), which is one of several untyped-success commands (`config-path`, `audit-open`, `skill-manifest`, and others). The routing oracle cannot assert on `smart-entry`'s routing decision until it gains a `--json` mode; that is future work, not covered here.
|
||||
- `smart-entry` emits prose unconditionally on success (`msd-core/bin/lib/smart-entry.cjs:577-587`), which is one of several untyped-success commands (`config-path`, `audit-open`, `skill-manifest`, and others). The routing oracle cannot assert on `smart-entry`'s routing decision until it gains a `--json` mode; that is future work, not covered here.
|
||||
- The open `docs/json-errors.md` vs. `src/cli-exit.cts` contract conflict (§6) remains unresolved by this ADR.
|
||||
|
||||
## References
|
||||
|
||||
- `CONTEXT.md` → `RULESET.TESTS.feedback-loop-convergence`
|
||||
- `tests/estimate-loop-convergence.test.cjs` — the existing convergence instance this ADR generalizes
|
||||
- `tests/helpers.cjs` — `runGsdTools`, the substrate the walk is layered over
|
||||
- `tests/helpers.cjs` — `runMsdTools`, the substrate the walk is layered over
|
||||
- `tests/qa/loop-walk.cjs`, `tests/qa/result.cjs`, `tests/qa/oracles.cjs`, `tests/qa/mutations.cjs`, `tests/qa/fixtures/**`, `tests/qa/scenarios/*.json`
|
||||
- `tests/loop-walk.qa.test.cjs`
|
||||
- `src/io.cts` — `output` (`output({error})` soft-failure idiom, 42 sites, `CRITICAL` per `get_impact`), `error`
|
||||
@@ -94,6 +94,6 @@ These are observations for a maintainer to weigh, not defects this PR fixes.
|
||||
- `src/cli-exit.cts` — `runMain`, the `ExitError` early-return that produces the §6 conflict
|
||||
- `docs/json-errors.md`
|
||||
- `CONTRIBUTING.md` → "Prohibited: Raw Text Matching on Test Outputs"
|
||||
- `gsd-core/bin/lib/smart-entry.cjs:577-587`
|
||||
- `msd-core/bin/lib/smart-entry.cjs:577-587`
|
||||
- Issue #2371 (fixture provenance standard)
|
||||
- `.gsd/phase/test-2966-loop-qa-walk/40-design.md` — the design this ADR records
|
||||
- `.msd/phase/test-2966-loop-qa-walk/40-design.md` — the design this ADR records
|
||||
|
||||
Reference in New Issue
Block a user