fix: allow publish when post-compaction prompt is within budget + remove reskill (#513)
* fix: allow publish when post-compaction prompt is within budget + remove reskill step - publish_eligible now depends only on prompt_within_budget, not degraded flag - Remove degraded-as-blocker from load_preflight() and assert_eligible_from_root() - Remove reskill-check and reskill jobs from crawl-and-publish workflow - Update tests to match new behavior Fixes #510 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * Add test for degraded=True + publish_eligible=True promotable case Addresses review comment on PR #513: covers the post-compaction within-budget scenario where degraded candidates remain promotable. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Juan Manuel Servera committed
Jun 16, 2026 at 19:35 UTC
1e7a23f7a94d38261bc62fe981b48902765c95f0
6 files changed
+41
-249
.github/workflows/crawl-and-publish.yml
-186
@@ -1350,189 +1350,3 @@ jobs:
1350
--body "$BODY"
1351
fi
1352
1353
- reskill-check:
1354
- needs: [crawl]
1355
- runs-on: ubuntu-latest
1356
- outputs:
1357
- should_reskill: ${{ steps.check.outputs.reskill }}
1358
-
1359
- steps:
1360
- - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
1361
- with:
1362
- fetch-depth: 0
1363
- ref: publish
1364
- persist-credentials: false
1365
- continue-on-error: true
1366
-
1367
- - name: Fallback to default branch for counter
1368
- env:
1369
- DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
1370
- run: |
1371
- if [ ! -f .squad/run-counter.txt ]; then
1372
- git fetch origin "$DEFAULT_BRANCH" 2>/dev/null || true
1373
- git checkout "origin/$DEFAULT_BRANCH" -- .squad/run-counter.txt 2>/dev/null || true
1374
- fi
1375
-
1376
- - name: Check reskill trigger
1377
- id: check
1378
- run: |
1379
- COUNTER=$(cat .squad/run-counter.txt 2>/dev/null || echo 0)
1380
- if [ $((COUNTER % 5)) -eq 0 ] && [ "$COUNTER" -gt 0 ]; then
1381
- echo "reskill=true" >> $GITHUB_OUTPUT
1382
- echo "🔄 Reskill triggered at run #$COUNTER"
1383
- else
1384
- echo "reskill=false" >> $GITHUB_OUTPUT
1385
- echo "📊 Run #$COUNTER — next reskill at run #$(( (COUNTER/5 + 1) * 5 ))"
1386
- fi
1387
-
1388
- reskill:
1389
- needs: [reskill-check]
1390
- if: needs.reskill-check.outputs.should_reskill == 'true'
1391
- runs-on: ubuntu-latest
1392
- permissions:
1393
- contents: write
1394
-
1395
- steps:
1396
- - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 # zizmor: ignore[artipacked] reskill job pushes to the publish branch; checkout token is reused by a later git push
1397
- with:
1398
- fetch-depth: 0
1399
- ref: publish
1400
- continue-on-error: true
1401
-
1402
- - name: Fallback to default branch
1403
- env:
1404
- DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
1405
- run: |
1406
- if [ ! -d .squad ]; then
1407
- git fetch origin "$DEFAULT_BRANCH"
1408
- git checkout "origin/$DEFAULT_BRANCH" -- . 2>/dev/null || true
1409
- fi
1410
-
1411
- - name: Set up Node
1412
- uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
1413
- with:
1414
- node-version: '24'
1415
-
1416
- - name: Set up Python
1417
- uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
1418
- with:
1419
- python-version: '3.12'
1420
-
1421
- - name: Install Copilot CLI
1422
- run: npm install -g @github/copilot
1423
-
1424
- - name: Run reskill
1425
- env:
1426
- DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
1427
- COPILOT_GITHUB_TOKEN: ${{ secrets.COPILOT_GH_TOKEN }}
1428
- GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
1429
- GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
1430
- run: |
1431
- set -euo pipefail
1432
- git config user.name "github-actions[bot]"
1433
- git config user.email "github-actions[bot]@users.noreply.github.com"
1434
- COUNTER=$(cat .squad/run-counter.txt 2>/dev/null || echo 0)
1435
- CURRENT_DATETIME=$(date -u +%Y-%m-%dT%H:%M:%SZ)
1436
- WEEK=$(date -u +%G-W%V)
1437
- RESKILL_OUTPUT=".squad/reskill/${WEEK}.md"
1438
- RESKILL_PROMPT=".squad/reskill/current-prompt.md"
1439
- mkdir -p .squad/skills .squad/reskill data/metrics
1440
-
1441
- sanitize_agent_output() {
1442
- python3 scripts/sanitize_agent_output.py --path "$1"
1443
- }
1444
-
1445
- # Primary path: Copilot CLI with agent identity
1446
- RESKILL_SOURCE="copilot-cli"
1447
- RESKILL_MODEL="copilot-default"
1448
- python3 scripts/reskill.py --current-datetime "$CURRENT_DATETIME" --output "$RESKILL_OUTPUT" --prompt-output "$RESKILL_PROMPT" --print-prompt > "$RESKILL_PROMPT" || true
1449
-
1450
- # Intentionally rely on Copilot CLI's default model so CI follows the platform-supported default.
1451
- RESKILL_FAILURE_CLASS=""
1452
- if command -v copilot >/dev/null 2>&1; then
1453
- set +e
1454
- copilot \
1455
- --agent weekly-analysis \
1456
- -p "Read the file at ${RESKILL_PROMPT}. Write the complete reskill markdown to ${RESKILL_OUTPUT}. The run is incomplete until ${RESKILL_OUTPUT} exists and is non-empty. Do not delegate. Do not spawn sub-agents. Do not emit commentary." \
1457
- -s \
1458
- --no-ask-user \
1459
- --allow-tool=read \
1460
- --allow-tool=write \
1461
- > /dev/null
1462
- COPILOT_STATUS=$?
1463
- set -e
1464
- if [ "$COPILOT_STATUS" -eq 0 ] && test -s "$RESKILL_OUTPUT"; then
1465
- echo "✅ Reskill via focused analysis agent"
1466
- else
1467
- if [ "$COPILOT_STATUS" -eq 0 ]; then
1468
- RESKILL_FAILURE_CLASS="writer_contract_failure"
1469
- echo "::warning::Reskill writer contract failed: ${RESKILL_OUTPUT} missing or empty after Copilot completed."
1470
- else
1471
- RESKILL_FAILURE_CLASS="copilot_cli_failure"
1472
- echo "::warning::Reskill Copilot CLI invocation failed with exit code ${COPILOT_STATUS}."
1473
- fi
1474
- RESKILL_SOURCE="none"
1475
- RESKILL_MODEL="none"
1476
- rm -f "$RESKILL_PROMPT"
1477
- echo "Copilot CLI reskill failed; no GitHub Models/OpenAI reskill fallback is configured. Writing placeholder trigger log."
1478
- {
1479
- echo "Reskill triggered at run #$COUNTER ($CURRENT_DATETIME)"
1480
- echo "Failure class: ${RESKILL_FAILURE_CLASS}"
1481
- } >> .squad/reskill/trigger-log.txt
1482
- fi
1483
- else
1484
- RESKILL_FAILURE_CLASS="copilot_inaccessible"
1485
- RESKILL_SOURCE="none"
1486
- RESKILL_MODEL="none"
1487
- rm -f "$RESKILL_PROMPT"
1488
- echo "Copilot CLI reskill failed; no GitHub Models/OpenAI reskill fallback is configured. Writing placeholder trigger log."
1489
- {
1490
- echo "Reskill triggered at run #$COUNTER ($CURRENT_DATETIME)"
1491
- echo "Failure class: ${RESKILL_FAILURE_CLASS}"
1492
- } >> .squad/reskill/trigger-log.txt
1493
- fi
1494
-
1495
- if [ "$RESKILL_SOURCE" != "none" ]; then
1496
- sanitize_agent_output "$RESKILL_OUTPUT"
1497
- fi
1498
-
1499
- if [ "$RESKILL_SOURCE" != "none" ]; then
1500
- API_RESPONSE_ARGS=""
1501
- if [ -f "data/metrics/reskill-api-response.json" ]; then
1502
- API_RESPONSE_ARGS="--api-response data/metrics/reskill-api-response.json"
1503
- fi
1504
- python3 scripts/track_token_usage.py \
1505
- --stage reskill \
1506
- --source "$RESKILL_SOURCE" \
1507
- --model "$RESKILL_MODEL" \
1508
- --current-datetime "$CURRENT_DATETIME" \
1509
- --week "$WEEK" \
1510
- --prompt-file "$RESKILL_PROMPT" \
1511
- --output-file "$RESKILL_OUTPUT" \
1512
- $API_RESPONSE_ARGS
1513
- rm -f "$RESKILL_PROMPT"
1514
- echo "🔄 Reskill report generated for run #$COUNTER"
1515
- fi
1516
-
1517
- if ! git status --short -- .squad data/metrics | grep -q .; then
1518
- echo "No .squad or token usage changes to commit."
1519
- exit 0
1520
- fi
1521
-
1522
- cp -r .squad squad-state-backup
1523
- cp -r data/metrics reskill-metrics-backup
1524
- DATA_BRANCH="publish"
1525
- if git fetch origin "$DATA_BRANCH" 2>/dev/null; then
1526
- git checkout -f -B "$DATA_BRANCH" "origin/$DATA_BRANCH"
1527
- else
1528
- git fetch origin "$DEFAULT_BRANCH"
1529
- git checkout -f -B "$DATA_BRANCH" "origin/$DEFAULT_BRANCH"
1530
- fi
1531
- cp -r squad-state-backup/* .squad/ 2>/dev/null || true
1532
- mkdir -p data/metrics
1533
- cp -r reskill-metrics-backup/* data/metrics/ 2>/dev/null || true
1534
- rm -rf squad-state-backup reskill-metrics-backup
1535
- git add .squad/ data/metrics/
1536
- git diff --cached --quiet && exit 0
1537
- git commit -m "chore: reskill state update [run #${GITHUB_RUN_ID}]"
1538
- git push origin "$DATA_BRANCH"
scripts/analyze_fallback.py
+3
-3
@@ -957,11 +957,11 @@ def _build_prompt(
957
},
958
prompt_within_budget=prompt_within_budget,
959
degraded=degraded,
960
- publish_eligible=prompt_within_budget and not degraded,
960
+ publish_eligible=prompt_within_budget,
961
promotion_policy=(
962
"normal-promotion"
963
- if not degraded
964
- else "staged/candidate-only by default; degraded compacted output requires an explicit future promotion policy."
963
+ if prompt_within_budget
964
+ else "staged/candidate-only by default; prompt exceeds token budget."
965
),
966
degradation_reason=degradation_reason,
967
fallback_policy=(
scripts/publish_manifest.py
-6
@@ -125,10 +125,6 @@ def load_preflight(path: Path | None, *, required: bool = False) -> tuple[dict[s
125
if payload is None:
126
return None, [f"preflight report missing or malformed: {path}"]
127
reasons: list[str] = []
128
- if payload.get("degraded") is True:
129
- reasons.append(
130
- "preflight degraded/compacted; candidate is staged-only unless an explicit promotion policy allows it"
131
- )
128
if payload.get("publish_eligible") is not True:
129
reasons.append("preflight report marks candidate as publish-ineligible")
130
return payload, reasons
@@ -715,8 +711,6 @@ def assert_eligible(args: argparse.Namespace) -> int:
711
preflight = analysis.get("preflight")
712
if not isinstance(preflight, dict) or preflight.get("publish_eligible") is not True:
713
raise SystemExit("Manifest lacks a publish-eligible Copilot preflight report.")
718
- if preflight.get("degraded") is True:
719
- raise SystemExit("Manifest preflight is degraded/compacted and staged-only by default.")
714
validation = payload.get("validation")
715
gate_report = validation.get("gate_report") if isinstance(validation, dict) else None
716
if not isinstance(gate_report, dict) or gate_report.get("present") is not True or gate_report.get("passed") is not True:
tests/test_analyze_fallback.py
+5
-4
@@ -412,13 +412,14 @@ class AnalyzeFallbackTests(unittest.TestCase):
412
report = json.loads(report_path.read_text(encoding="utf-8"))
413
self.assertEqual(exit_code, 0)
414
self.assertTrue(report["degraded"])
415
- self.assertFalse(report["publish_eligible"])
416
- self.assertIn("staged/candidate-only", report["promotion_policy"])
415
+ # Post-compaction prompt is within budget, so publish is eligible
416
+ self.assertTrue(report["publish_eligible"])
417
+ self.assertEqual("normal-promotion", report["promotion_policy"])
418
self.assertIn("compacted", report["degradation_reason"])
419
report_markdown = report_md_path.read_text(encoding="utf-8")
420
self.assertIn("Degraded/compacted: `true`", report_markdown)
420
- self.assertIn("Publish eligible: `false`", report_markdown)
421
- self.assertIn("staged/candidate-only", report_markdown)
421
+ self.assertIn("Publish eligible: `true`", report_markdown)
422
+ self.assertIn("normal-promotion", report_markdown)
423
components = {component["name"]: component for component in report["components"]}
424
self.assertIn("compacted to top", components["new_repos"]["compaction_decision"])
425
self.assertIn("compacted to top", components["trending_repos"]["compaction_decision"])
tests/test_pipeline.py
+5
-49
@@ -231,57 +231,13 @@ class WorkflowConfigTests(unittest.TestCase):
231
self.assertIn("UNTIL=$(date -u +%Y-%m-%d)", run_script)
232
self.assertIn('--until "$UNTIL"', run_script)
233
234
- def test_crawl_workflow_defines_reskill_jobs(self) -> None:
234
+ def test_crawl_workflow_defines_analyze_job(self) -> None:
235
workflow_path = Path(".github/workflows/crawl-and-publish.yml")
236
workflow = yaml.safe_load(workflow_path.read_text(encoding="utf-8"))
237
-
238
- self.assertIn("reskill-check", workflow["jobs"])
239
- reskill_check = workflow["jobs"]["reskill-check"]
240
- self.assertEqual(reskill_check["needs"], ["crawl"])
241
-
242
- self.assertIn("reskill", workflow["jobs"])
243
- reskill = workflow["jobs"]["reskill"]
244
- self.assertEqual(reskill["needs"], ["reskill-check"])
245
- self.assertIn("needs.reskill-check.outputs.should_reskill", reskill["if"])
246
-
247
- check_step = next((s for s in reskill_check["steps"] if s.get("name") == "Check reskill trigger"), None)
248
- self.assertIsNotNone(check_step)
249
- check_run = check_step["run"]
250
- self.assertIn("reskill=true", check_run)
251
- self.assertIn("$GITHUB_OUTPUT", check_run)
252
-
253
- install_step = next((s for s in reskill["steps"] if s.get("name") == "Install Copilot CLI"), None)
254
- self.assertIsNotNone(install_step)
255
- self.assertEqual(install_step["run"], "npm install -g @github/copilot")
256
- self.assertNotIn("continue-on-error", install_step)
257
-
258
- reskill_step = next((s for s in reskill["steps"] if s.get("name") == "Run reskill"), None)
259
- self.assertIsNotNone(reskill_step)
260
- self.assertNotIn("GITHUB_MODELS_MODEL", workflow["env"])
261
- self.assertEqual(reskill_step["env"]["COPILOT_GITHUB_TOKEN"], "${{ secrets.COPILOT_GH_TOKEN }}")
262
- reskill_run = reskill_step["run"]
263
- self.assertIn("python3 scripts/reskill.py --current-datetime", reskill_run)
264
- self.assertIn("--prompt-output", reskill_run)
265
- self.assertIn("python3 scripts/track_token_usage.py", reskill_run)
266
- self.assertIn('RESKILL_MODEL="copilot-default"', reskill_run)
267
- self.assertNotIn("--model claude-sonnet-4", reskill_run)
268
- self.assertIn("mkdir -p .squad/skills .squad/reskill", reskill_run)
269
- self.assertIn("data/metrics", reskill_run)
270
- self.assertIn("trigger-log.txt", reskill_run)
271
- self.assertIn("git add .squad/", reskill_run)
272
- self.assertIn("data/metrics/", reskill_run)
273
- self.assertIn("no GitHub Models/OpenAI reskill fallback", reskill_run)
274
- self.assertNotIn("${GITHUB_MODELS_MODEL}", reskill_run)
275
- self.assertNotIn('RESKILL_SOURCE="github-models"', reskill_run)
276
- self.assertNotIn("used GitHub Models API fallback", reskill_run)
277
- self.assertIn("--agent weekly-analysis", reskill_run)
278
- self.assertIn('Read the file at ${RESKILL_PROMPT}. Write the complete reskill markdown to ${RESKILL_OUTPUT}.', reskill_run)
279
- self.assertIn('test -s "$RESKILL_OUTPUT"', reskill_run)
280
- self.assertIn('RESKILL_FAILURE_CLASS="writer_contract_failure"', reskill_run)
281
- self.assertNotIn("--allow-tool=glob", reskill_run)
282
- self.assertNotIn("--allow-tool=grep", reskill_run)
283
- # Prompt is written to a well-known path, not a temp file
284
- self.assertIn('RESKILL_PROMPT=".squad/reskill/current-prompt.md"', reskill_run)
237
+
238
+ # Reskill jobs have been removed (analysis uses plain copilot-cli)
239
+ self.assertNotIn("reskill-check", workflow["jobs"])
240
+ self.assertNotIn("reskill", workflow["jobs"])
241
242
analyze = workflow["jobs"]["analyze"]
243
preflight_step = next((s for s in analyze["steps"] if s.get("name") == "Render and preflight analysis prompt"), None)
tests/test_publish_manifest.py
+28
-1
@@ -313,10 +313,37 @@ class PublishManifestTests(unittest.TestCase):
313
self.assertTrue(payload["analysis"]["preflight"]["degraded"])
314
self.assertFalse(payload["analysis"]["preflight"]["publish_eligible"])
315
self.assertIn("staged/candidate-only", payload["analysis"]["preflight"]["promotion_policy"])
316
- self.assertTrue(any("preflight degraded/compacted" in reason for reason in payload["promotion"]["reasons"]))
316
+ self.assertTrue(any("publish-ineligible" in reason for reason in payload["promotion"]["reasons"]))
317
with self.assertRaises(SystemExit):
318
assert_eligible_from_root(base, manifest)
319
320
+ def test_degraded_but_publish_eligible_candidate_is_promotable(self) -> None:
321
+ """Post-compaction within budget: degraded=True + publish_eligible=True is promotable."""
322
+ tests_root = Path(__file__).resolve().parent
323
+ with tempfile.TemporaryDirectory(dir=tests_root) as tmpdir:
324
+ base = Path(tmpdir)
325
+ raw = base / "data/raw/2026-W21.json"
326
+ summary = base / "data/candidates/2026-W21/123456/2026-W21-summary.md"
327
+ manifest = base / "data/candidates/2026-W21/123456/publish-manifest.json"
328
+ gate_report = base / "data/candidates/2026-W21/123456/analysis-gate-report.json"
329
+ preflight = base / "data/candidates/2026-W21/123456/diagnostics/analysis-preflight.json"
330
+ write_raw(raw)
331
+ write_summary(summary)
332
+ write_gate_report(gate_report)
333
+ write_preflight(preflight, degraded=True, publish_eligible=True)
334
+
335
+ publish_manifest.main(
336
+ create_args(base, raw, summary, manifest, gate_report=gate_report, preflight=preflight)
337
+ )
338
+
339
+ payload = json.loads(manifest.read_text(encoding="utf-8"))
340
+ self.assertTrue(payload["promotion"]["eligible"])
341
+ self.assertEqual(payload["promotion"]["decision"], "promote")
342
+ self.assertTrue(payload["analysis"]["preflight"]["degraded"])
343
+ self.assertTrue(payload["analysis"]["preflight"]["publish_eligible"])
344
+ # Should not raise — degraded but eligible means promotable
345
+ assert_eligible_from_root(base, manifest)
346
+
347
def test_copilot_candidate_without_explicit_model_uses_publishable_default(self) -> None:
348
tests_root = Path(__file__).resolve().parent
349
with tempfile.TemporaryDirectory(dir=tests_root) as tmpdir: