rungate 0.2.1 → 0.4.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/bin/rungate.ts +70 -0
- package/lib/aes-calculator.ts +66 -0
- package/lib/agent-audit.ts +139 -0
- package/lib/agent-brief-validation.ts +359 -0
- package/lib/agents-md-refs.ts +90 -0
- package/lib/bash-file-read.ts +197 -0
- package/lib/behavioral-cache.ts +78 -0
- package/lib/branch-cleanup.ts +234 -0
- package/lib/brief-context-parser.ts +35 -0
- package/lib/brief-precedence.ts +114 -0
- package/lib/canary.ts +117 -0
- package/lib/check-registry.ts +226 -0
- package/lib/ci-checkout.ts +426 -0
- package/lib/cli-commands.ts +137 -0
- package/lib/compliance-backfill.ts +232 -0
- package/lib/compliance-report.ts +605 -0
- package/lib/compliance.ts +361 -0
- package/lib/config-loader.ts +255 -0
- package/lib/config-readers.ts +86 -0
- package/lib/conformity.ts +2161 -137
- package/lib/container-provenance.ts +230 -0
- package/lib/create-brief.ts +150 -0
- package/lib/create-sc.ts +267 -0
- package/lib/decomposition-waves.ts +197 -0
- package/lib/deep-merge.ts +69 -0
- package/lib/dev-liveness.ts +214 -0
- package/lib/directive-extractor.ts +210 -0
- package/lib/distribution-standing.ts +113 -0
- package/lib/doc-hygiene.ts +238 -0
- package/lib/eval-criteria.ts +6 -0
- package/lib/evidence-portability.ts +123 -0
- package/lib/evidence-prevalidator.ts +390 -0
- package/lib/evidence-required.ts +150 -0
- package/lib/evidence-satisfiability.ts +288 -0
- package/lib/frontmatter.ts +163 -0
- package/lib/gap-scanner.ts +348 -0
- package/lib/gate-check-audit.ts +190 -0
- package/lib/gate-enforcement.ts +366 -0
- package/lib/generators/agent-briefs.ts +136 -0
- package/lib/generators/agents-md.ts +511 -0
- package/lib/generators/code-map.ts +60 -0
- package/lib/generators/types.ts +184 -0
- package/lib/git-ref-normalizer.ts +107 -0
- package/lib/github.ts +539 -0
- package/lib/harness-identity.ts +63 -0
- package/lib/hill-climb.ts +164 -0
- package/lib/instruction-effectiveness.ts +131 -0
- package/lib/issue-closure-audit.ts +225 -0
- package/lib/issue-create-override.ts +343 -0
- package/lib/issue-extraction.ts +368 -0
- package/lib/js-top-level-code.ts +250 -0
- package/lib/liveness-contradiction.ts +149 -0
- package/lib/organize.ts +505 -0
- package/lib/paths.ts +197 -6
- package/lib/phase2-decomposition.ts +270 -0
- package/lib/post-fix-verify.ts +157 -0
- package/lib/prior-branch.ts +512 -0
- package/lib/project-state-shape.ts +166 -0
- package/lib/promote-outputs.ts +156 -0
- package/lib/prove-label-gate.ts +128 -0
- package/lib/reachability.ts +302 -0
- package/lib/release-preflight.ts +288 -0
- package/lib/rule-health-pipeline.ts +264 -0
- package/lib/rule-health-trend.ts +161 -0
- package/lib/rule-provenance.ts +149 -0
- package/lib/rule-registry.ts +128 -0
- package/lib/rungate-schema.ts +127 -5
- package/lib/sc-claim-forms.ts +179 -0
- package/lib/sc-cover.ts +150 -0
- package/lib/sc-guard.ts +157 -0
- package/lib/sc-ids.ts +344 -0
- package/lib/scaffold/defaults.ts +165 -0
- package/lib/scaffold/managed-workflow.ts +370 -0
- package/lib/scaffold/owned-files.ts +265 -0
- package/lib/scaffold/pre-push-hook.ts +165 -0
- package/lib/scaffold/steps.ts +2621 -0
- package/lib/scanner.ts +701 -0
- package/lib/scope-denylist.ts +258 -0
- package/lib/screenshot-upload.ts +73 -0
- package/lib/secret-patterns.ts +112 -0
- package/lib/security-verdict.ts +616 -0
- package/lib/signal-phrases.ts +1 -0
- package/lib/spec-change-conformity.ts +80 -0
- package/lib/spec-files.ts +62 -0
- package/lib/spec-registry.ts +133 -0
- package/lib/spec-updater.ts +96 -0
- package/lib/stale-cleanup.ts +224 -0
- package/lib/stale-issue-scanner.ts +36 -0
- package/lib/strip-js-comments.ts +269 -0
- package/lib/suite-measurement.ts +228 -0
- package/lib/task-completion-checks.ts +224 -0
- package/lib/test-suite-lock.ts +1136 -0
- package/lib/transcript-checker.ts +1140 -0
- package/lib/typescript-project.ts +233 -0
- package/lib/validators/spec-validators.ts +171 -0
- package/lib/verdict-capture.ts +73 -0
- package/lib/verified-push.ts +535 -0
- package/lib/workflow-security.ts +174 -0
- package/lib/worktree-cleanup.ts +432 -0
- package/lib/worktree-collect.ts +283 -0
- package/lib/worktree-isolation.ts +82 -0
- package/lib/worktree-repo.ts +300 -0
- package/package.json +27 -3
- package/prompts/ac-adversary.md +14 -2
- package/prompts/aditi.md +5 -0
- package/prompts/blast-radius.md +5 -0
- package/prompts/container-verify.md +5 -0
- package/prompts/discovery.md +4 -0
- package/prompts/environment.md +5 -0
- package/prompts/evidence-hierarchy.md +5 -0
- package/prompts/evidence-validator.md +11 -0
- package/prompts/marcus.md +5 -0
- package/prompts/prevention.md +5 -0
- package/prompts/prove-reproducer.md +12 -0
- package/prompts/quinn-decision-tree.md +5 -0
- package/prompts/quinn-ui-brief.md +5 -0
- package/prompts/quinn.md +5 -0
- package/prompts/rca.md +5 -0
- package/prompts/read-before-write.md +4 -0
- package/prompts/regression.md +5 -0
- package/prompts/serena.md +5 -0
- package/scripts/audit-specs.ts +464 -0
- package/scripts/convert-spec.ts +293 -0
- package/scripts/create-sc.ts +92 -0
- package/scripts/create-spec.ts +166 -0
- package/scripts/extract-constraints.ts +4 -3
- package/scripts/generate-code-map.ts +1 -1
- package/scripts/scaffold-project.ts +138 -1426
- package/scripts/sync-spec-tests.ts +211 -126
package/package.json
CHANGED
|
@@ -1,31 +1,55 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "rungate",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.4.0",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"description": "Ship harness — conformity tests, scaffold, and agent briefs for AI-first development",
|
|
6
|
+
"repository": {
|
|
7
|
+
"type": "git",
|
|
8
|
+
"url": "git+https://github.com/hornjason/pai-harness.git"
|
|
9
|
+
},
|
|
10
|
+
"homepage": "https://github.com/hornjason/pai-harness#readme",
|
|
11
|
+
"bugs": {
|
|
12
|
+
"url": "https://github.com/hornjason/pai-harness/issues"
|
|
13
|
+
},
|
|
6
14
|
"files": [
|
|
15
|
+
"bin/",
|
|
7
16
|
"lib/",
|
|
8
17
|
"prompts/",
|
|
9
18
|
"scripts/scaffold-project.ts",
|
|
10
19
|
"scripts/extract-constraints.ts",
|
|
20
|
+
"scripts/convert-spec.ts",
|
|
11
21
|
"scripts/sync-spec-tests.ts",
|
|
12
22
|
"scripts/generate-code-map.ts",
|
|
23
|
+
"scripts/audit-specs.ts",
|
|
24
|
+
"scripts/create-spec.ts",
|
|
25
|
+
"scripts/create-sc.ts",
|
|
13
26
|
"package.json"
|
|
14
27
|
],
|
|
15
28
|
"exports": {
|
|
16
29
|
"./lib/conformity": "./lib/conformity.ts",
|
|
17
30
|
"./lib/rungate-schema": "./lib/rungate-schema.ts",
|
|
18
|
-
"./lib/signal-phrases": "./lib/signal-phrases.ts"
|
|
31
|
+
"./lib/signal-phrases": "./lib/signal-phrases.ts",
|
|
32
|
+
"./lib/compliance": "./lib/compliance.ts",
|
|
33
|
+
"./scripts/audit-specs": "./scripts/audit-specs.ts"
|
|
19
34
|
},
|
|
20
35
|
"scripts": {
|
|
21
36
|
"test": "bun test",
|
|
37
|
+
"posttest": "bun scripts/sync-sc-status.ts",
|
|
38
|
+
"sync-scs": "bun scripts/sync-sc-status.ts",
|
|
39
|
+
"typecheck": "bun scripts/typecheck.ts",
|
|
22
40
|
"test:structure": "bun test test/structure.test.ts",
|
|
23
41
|
"test:external-deps": "bun test test/external-deps.test.ts"
|
|
24
42
|
},
|
|
25
43
|
"dependencies": {
|
|
44
|
+
"@octokit/rest": "^22.0.1",
|
|
45
|
+
"acorn": "^8.19.0",
|
|
26
46
|
"zod": "^3.23.0"
|
|
27
47
|
},
|
|
28
48
|
"devDependencies": {
|
|
29
|
-
"@types/bun": "latest"
|
|
49
|
+
"@types/bun": "latest",
|
|
50
|
+
"typescript": "5.9.3"
|
|
51
|
+
},
|
|
52
|
+
"bin": {
|
|
53
|
+
"rungate": "bin/rungate.ts"
|
|
30
54
|
}
|
|
31
55
|
}
|
package/prompts/ac-adversary.md
CHANGED
|
@@ -1,3 +1,10 @@
|
|
|
1
|
+
---
|
|
2
|
+
doc-type: reference
|
|
3
|
+
status: active
|
|
4
|
+
owner: jason
|
|
5
|
+
updated: 2026-09-20
|
|
6
|
+
---
|
|
7
|
+
|
|
1
8
|
You are an adversarial AC reviewer. Your objective: find ways to pass
|
|
2
9
|
every AC without actually fixing the bug.
|
|
3
10
|
|
|
@@ -10,11 +17,16 @@ For each AC, answer:
|
|
|
10
17
|
|
|
11
18
|
Output as JSON:
|
|
12
19
|
{
|
|
13
|
-
"gameable":
|
|
14
|
-
"approved":
|
|
20
|
+
"gameable": NUMBER_OF_GAMEABLE_ACS,
|
|
21
|
+
"approved": true_IF_0_GAMEABLE,
|
|
15
22
|
"exploits": [
|
|
16
23
|
{"acId": "SC-1", "exploit": "how to pass without fixing", "recommendation": "how to tighten"}
|
|
17
24
|
]
|
|
18
25
|
}
|
|
19
26
|
|
|
20
27
|
If ALL ACs are robust, output: {"gameable": 0, "approved": true, "exploits": []}
|
|
28
|
+
|
|
29
|
+
### Never
|
|
30
|
+
- Never approve ACs where the evidence command can be passed by code that doesn't fix the stated problem
|
|
31
|
+
- Never accept self-attestation ("I verified it works") as a valid evidence method
|
|
32
|
+
- Never rate a structural check (file exists, grep count) as equivalent to a behavioral check (runtime output)
|
package/prompts/aditi.md
CHANGED
|
@@ -27,3 +27,8 @@ You are Aditi Sharma, UI/UX designer. Specify component layouts and interaction
|
|
|
27
27
|
### Constraints
|
|
28
28
|
- Use existing design system components first
|
|
29
29
|
- shadcn/ui as component library baseline
|
|
30
|
+
|
|
31
|
+
### Never
|
|
32
|
+
- Never skip accessibility requirements — every component spec must include ARIA roles and keyboard navigation
|
|
33
|
+
- Never ignore existing design system components in favor of custom elements — check shadcn/ui first
|
|
34
|
+
- Never deliver a component spec without error states and loading states
|
package/prompts/blast-radius.md
CHANGED
|
@@ -27,3 +27,8 @@ filesRead >= filesChanged — you must read more than you change.
|
|
|
27
27
|
### Gate enforcement
|
|
28
28
|
- Read-before-write ratio >= 3:1 (read tokens / write tokens)
|
|
29
29
|
- Files changed outside brief's listed files = WARN
|
|
30
|
+
|
|
31
|
+
### Never
|
|
32
|
+
- Never change files not listed in the brief without explicit scope expansion approval
|
|
33
|
+
- Never skip reading dependent files before modifying their imports — downstream breakage is the most common blast-radius failure
|
|
34
|
+
- Never proceed when filesChanged > filesRead — stop and read more
|
|
@@ -26,3 +26,8 @@ After container-rebuild completes — before declaring deploy success.
|
|
|
26
26
|
- Container HEAD: {sha}
|
|
27
27
|
- Health: {response}
|
|
28
28
|
- Failures: [list or "none"]
|
|
29
|
+
|
|
30
|
+
### Never
|
|
31
|
+
- Never skip the health endpoint check — a running container is not the same as a healthy container
|
|
32
|
+
- Never declare container verified without completing smoke tests on critical endpoints
|
|
33
|
+
- Never report "verified" when the container HEAD doesn't match the expected commit SHA
|
package/prompts/discovery.md
CHANGED
|
@@ -25,3 +25,7 @@ DISCOVERY phase of ship cycle — before writing any code.
|
|
|
25
25
|
- Key constraints discovered
|
|
26
26
|
- Files that will be touched
|
|
27
27
|
- Risks or unknowns to flag
|
|
28
|
+
|
|
29
|
+
### Never
|
|
30
|
+
- Never write code before completing the discovery read list — implementation without context produces scope violations
|
|
31
|
+
- Never skip reading AGENTS.md — it contains project identity, constraints, and commands that govern all work
|
package/prompts/environment.md
CHANGED
|
@@ -26,3 +26,8 @@ Ship DISCOVERY phase — after reading docs, before coding.
|
|
|
26
26
|
- Environment: ready / blocked
|
|
27
27
|
- Baseline test results: N pass, N fail
|
|
28
28
|
- Blockers: [list or "none"]
|
|
29
|
+
|
|
30
|
+
### Never
|
|
31
|
+
- Never skip environment verification and proceed directly to coding — a broken baseline wastes the entire session
|
|
32
|
+
- Never assume dependencies are installed — run the checks and capture actual output
|
|
33
|
+
- Never report "ready" with failing type checks or missing env vars
|
|
@@ -29,3 +29,8 @@ Gate enforcement — maps evidence types to tiers per AC type.
|
|
|
29
29
|
- UI → B (browser verification)
|
|
30
30
|
- BUG-FIX → S (negative control)
|
|
31
31
|
- Static analysis → C supplementary only
|
|
32
|
+
|
|
33
|
+
### Never
|
|
34
|
+
- Never accept tier F (self-attestation) evidence for any AC — it is always rejected
|
|
35
|
+
- Never mix evidence tiers to inflate a verdict — the weakest piece of evidence governs the tier
|
|
36
|
+
- Never use static grep (tier D) as the sole evidence for more than 25% of ACs
|
|
@@ -1,3 +1,10 @@
|
|
|
1
|
+
---
|
|
2
|
+
doc-type: reference
|
|
3
|
+
status: active
|
|
4
|
+
owner: jason
|
|
5
|
+
updated: 2026-09-20
|
|
6
|
+
---
|
|
7
|
+
|
|
1
8
|
You are an evidence validator. Your objective: run evidence commands
|
|
2
9
|
independently and report whether results meet AC thresholds.
|
|
3
10
|
|
|
@@ -20,3 +27,7 @@ Report as JSON:
|
|
|
20
27
|
|
|
21
28
|
Do not infer, assume, or extrapolate. If the command fails to run,
|
|
22
29
|
verdict is FAIL with the error message as rawOutput.
|
|
30
|
+
|
|
31
|
+
### Never
|
|
32
|
+
- Never infer or extrapolate evidence from partial command output — if the command didn't produce a clear result, verdict is FAIL
|
|
33
|
+
- Never modify source code — you run in a read-only worktree and exist solely to validate evidence
|
package/prompts/marcus.md
CHANGED
|
@@ -27,3 +27,8 @@ You are Marcus Webb, principal engineer. You implement code changes, write tests
|
|
|
27
27
|
|
|
28
28
|
### Report format
|
|
29
29
|
For each AC, provide evidence: file:line, grep output, or test result.
|
|
30
|
+
|
|
31
|
+
### Never
|
|
32
|
+
- Never skip reading AGENTS.md as the first action — it contains project constraints that govern implementation
|
|
33
|
+
- Never report done without running the full test suite and capturing pass/fail counts
|
|
34
|
+
- Never implement without writing a failing test first (TDD is mandatory)
|
package/prompts/prevention.md
CHANGED
|
@@ -24,3 +24,8 @@ After identifying root cause — before implementing the fix.
|
|
|
24
24
|
### Example
|
|
25
25
|
Bad: `if (x) doThing(x)` — fixes one caller
|
|
26
26
|
Good: `function doThing(x: NonNullable<T>)` — prevents all callers
|
|
27
|
+
|
|
28
|
+
### Never
|
|
29
|
+
- Never fix only the immediate instance without auditing sibling files for the same pattern
|
|
30
|
+
- Never skip the regression test — a fix without a test that would have caught it is incomplete
|
|
31
|
+
- Never add a guard without narrowing the type — runtime checks that the compiler can't enforce will recur
|
|
@@ -1,3 +1,10 @@
|
|
|
1
|
+
---
|
|
2
|
+
doc-type: reference
|
|
3
|
+
status: active
|
|
4
|
+
owner: jason
|
|
5
|
+
updated: 2026-09-20
|
|
6
|
+
---
|
|
7
|
+
|
|
1
8
|
You are a prove reproducer. Your objective: reproduce the bug described
|
|
2
9
|
in the issue, verify the fix works, and capture evidence.
|
|
3
10
|
|
|
@@ -41,3 +48,8 @@ Verdict rules:
|
|
|
41
48
|
- PROVEN: bug was reproduced in before-state AND fix verified in after-state
|
|
42
49
|
- UNPROVEN: fix does not resolve the issue (evidence shows problem persists)
|
|
43
50
|
- INCONCLUSIVE: could not reproduce the bug or connect to dev server
|
|
51
|
+
|
|
52
|
+
### Never
|
|
53
|
+
- Never claim UI verification without browser tools — output SKIP with reason "no browser tools — requires Quinn"
|
|
54
|
+
- Never infer evidence — every verdict needs captured command output, API response, or grep result
|
|
55
|
+
- Never report PROVEN without both reproducing the bug in before-state AND verifying the fix in after-state
|
|
@@ -29,3 +29,8 @@ Embedded in Quinn's brief — referenced during journey execution.
|
|
|
29
29
|
### Circuit breaker
|
|
30
30
|
3 consecutive FAIL steps → abort journey, report partial results.
|
|
31
31
|
Max 8 steps per journey — split longer flows.
|
|
32
|
+
|
|
33
|
+
### Never
|
|
34
|
+
- Never skip pre-conditions — if the page isn't loaded or shows errors, report FAIL immediately without testing ACs
|
|
35
|
+
- Never report PASS without a11y snapshot or screenshot evidence backing the assertion
|
|
36
|
+
- Never use screenshots as the primary assertion method — browser_snapshot (a11y tree) is always primary
|
|
@@ -93,6 +93,11 @@ Overall: PASS | FAIL
|
|
|
93
93
|
{If FAIL: blocking issues listed}
|
|
94
94
|
```
|
|
95
95
|
|
|
96
|
+
### Never
|
|
97
|
+
- Never guess URLs — always use the pages map from rungate.json for target URLs
|
|
98
|
+
- Never skip viewport setup (browser_resize) before testing — assertions against wrong viewport produce false passes
|
|
99
|
+
- Never report PASS without attaching snapshot or screenshot evidence for each AC
|
|
100
|
+
|
|
96
101
|
## Interpolation Variables
|
|
97
102
|
|
|
98
103
|
| Variable | Source |
|
package/prompts/quinn.md
CHANGED
|
@@ -27,3 +27,8 @@ Each step: action → wait_for → assertion → on_fail
|
|
|
27
27
|
|
|
28
28
|
### Circuit breaker
|
|
29
29
|
3 consecutive FAIL steps → abort journey with partial results.
|
|
30
|
+
|
|
31
|
+
### Never
|
|
32
|
+
- Never trust prior test results — always run fresh browser assertions in the current state
|
|
33
|
+
- Never skip anti-checks even when all AC journey steps pass
|
|
34
|
+
- Never report PASS without snapshot or screenshot evidence attached
|
package/prompts/rca.md
CHANGED
|
@@ -26,3 +26,8 @@ Marcus brief for bug-fix issues — required RCA section.
|
|
|
26
26
|
3. Identify the exact line where behavior diverges from expected
|
|
27
27
|
4. State root cause as "X happens because Y at file:line"
|
|
28
28
|
5. Predict the fix outcome before implementing
|
|
29
|
+
|
|
30
|
+
### Never
|
|
31
|
+
- Never implement a fix without first reproducing the bug — an unreproduced bug means you're guessing at root cause
|
|
32
|
+
- Never skip the prediction step — state what will happen after the fix before implementing it
|
|
33
|
+
- Never fix at the symptom level when the root cause is upstream in the data flow
|
|
@@ -28,3 +28,7 @@ Read-to-write ratio >= 3:1 (measured in tokens).
|
|
|
28
28
|
|
|
29
29
|
### Gate check
|
|
30
30
|
`readTokens / writeTokens >= 3.0` — FAIL if under.
|
|
31
|
+
|
|
32
|
+
### Never
|
|
33
|
+
- Never write code before reading at least 3x the tokens you plan to write — the gate enforces this mechanically
|
|
34
|
+
- Never skip reading dependency files that import the files you'll change — downstream breakage from unread dependents is the top failure mode
|
package/prompts/regression.md
CHANGED
|
@@ -24,3 +24,8 @@ After implementing a fix — before marking AC complete.
|
|
|
24
24
|
### Gate enforcement
|
|
25
25
|
- Test count post >= test count pre — decrease = FAIL
|
|
26
26
|
- Zero test removals without explicit rationale
|
|
27
|
+
|
|
28
|
+
### Never
|
|
29
|
+
- Never delete existing tests to make a new fix pass — test removals require explicit rationale
|
|
30
|
+
- Never skip the negative control (reverting the fix to confirm the test fails without it)
|
|
31
|
+
- Never write a test that covers only the symptom — test the root cause
|
package/prompts/serena.md
CHANGED
|
@@ -26,3 +26,8 @@ You are Serena Blackwood, architect. Evaluate structural decisions before implem
|
|
|
26
26
|
|
|
27
27
|
### Output
|
|
28
28
|
ADR recommendation or architecture approval with cited file:line evidence.
|
|
29
|
+
|
|
30
|
+
### Never
|
|
31
|
+
- Never approve shallow wrappers that add indirection without information hiding — demand deep modules
|
|
32
|
+
- Never skip contract test coverage for shared interfaces — untested contracts break silently at integration
|
|
33
|
+
- Never approve an architectural change without citing file:line evidence from the current codebase
|