@fro.bot/systematic 3.12.3 → 3.12.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fro.bot/systematic",
-  "version": "3.12.3",
+  "version": "3.12.4",
   "description": "Compound-engineering loops for OpenCode, Pi, and Claude Code",
   "type": "module",
   "homepage": "https://fro.bot/systematic",

package/skills/ce-review/SKILL.md

@@ -456,17 +456,17 @@ The parent-owned artifact and its reconciliation rules are defined in the [synth
 
 Convert multiple reviewer JSON returns into one deduplicated, confidence-gated finding set. Each persona return already contains both tiers. The parent must retain the validated payload in memory for merge and synthesis, then persist only the same validated data.
 
-Before applying the confidence gate, assign every finding in a valid return a stable parent-owned `input_id` of `<reviewer>#<1-based finding index>` and keep the parent-owned ledger through every later stage. See the [synthesis artifact contract](./references/synthesis-artifact-contract.md) for the ledger's authoritative reconciliation rules.
+Before applying the confidence gate, keep the parent-owned ledger through every later stage. See the [synthesis artifact contract](./references/synthesis-artifact-contract.md) for the input-ID and reconciliation rules.
 
 1. **Validate before any write.** Treat every persona return as untrusted input. Parse the returned text as JSON without logging the raw text, then validate the complete parsed object against `references/findings-schema.json`, including `why_it_matters` and `evidence`.
    - **Top-level required:** reviewer (string), findings (array), residual_risks (array), testing_gaps (array). Reject the entire persona return if any are missing or wrong type.
    - **Per-finding required:** title, severity, file, line, why_it_matters, confidence, evidence, autofix_class, owner, requires_verification, pre_existing.
    - **Schema constraints:** enforce every enum, type, confidence, line, path, evidence count, evidence length, and explicit overflow-marker bound from the schema. Empty evidence, absolute paths, and over-bound evidence are rejection cases, not truncation cases.
-   - **Environment-value detection:** JSON Schema cannot determine where a string came from, so recursively inspect every string leaf in the parsed payload before writing. Reject the payload when a string contains a shell/environment reference (`$NAME`, `${NAME}`, `process.env.NAME`, or `os.environ[...]`), an assignment using a known environment variable (`NAME=value`), or a current environment value as an exact or embedded match. Use a conservative match set of non-empty runtime environment values; never log the matched value. This detector is an additional parent-side check, not a schema claim.
-   - **Safe rejection message:** report only `Rejected persona <name> return: field <JSON path> failed <reason>.` Derive `<JSON path>` from the validator or recursive scan and use a fixed reason such as `schema validation`, `environment-value detection`, or `malformed JSON`; never include the offending value, raw return, or validator parameters.
-   - **No partial writes:** do not write a per-agent record or merge any finding until the entire persona payload passes schema and environment-value validation. A valid payload is then annotated by the parent with `harness` and `dispatch_outcome` and written by the parent only. Revalidate the enriched record before persistence.
+   - **Environment-value detection:** JSON Schema cannot determine where a string came from, so recursively inspect every string leaf in the parsed payload before writing. Apply the environment-value matching, structural-detector, and finding-granularity rules in the [synthesis artifact contract](./references/synthesis-artifact-contract.md). This detector is an additional parent-side check, not a schema claim.
+   - **Safe rejection message:** Use the safe rejection message rule in the [synthesis artifact contract](./references/synthesis-artifact-contract.md); never include the offending value, raw return, or validator parameters.
+   - **No partial writes:** Do not write or merge a finding until it passes the parent-side validation rules. Apply the admitted-finding persistence and rejected-payload ledger rules in the [synthesis artifact contract](./references/synthesis-artifact-contract.md). A valid admitted finding is then annotated by the parent with `harness` and `dispatch_outcome` and written by the parent only. Revalidate the enriched record before persistence.
    - **Dispatch outcome:** Record the parent-owned dispatch outcomes and ledger dispositions according to the [synthesis artifact contract](./references/synthesis-artifact-contract.md); keep dispatch outcomes separate from finding dispositions.
-   - **Rejection policy: degrade, do not fail the whole review.** Continue merging conforming returns when a persona is rejected; record the rejection according to the [synthesis artifact contract](./references/synthesis-artifact-contract.md). If every persona fails or times out, use the existing degraded-review behavior.
+   - **Rejection policy: degrade, do not fail the whole review.** Continue merging conforming returns when a persona or finding is rejected; record the rejection and apply the risk-aware verdict according to the [synthesis artifact contract](./references/synthesis-artifact-contract.md). If every persona fails or times out, use the existing degraded-review behavior.
 2. **Confidence gate.** Suppress findings below 0.60 confidence. Exception: P0 findings at 0.50+ confidence survive the gate -- critical-but-uncertain issues must not be silently dropped. Record the suppressed finding's original confidence and an explicit reason in the input ledger. A retained P0 at 0.50+ is recorded as `surviving` unless it later participates in a deduplication merge. This matches the persona instructions and the schema's confidence thresholds.
 3. **Deduplicate.** Compute fingerprint: `normalize(file) + line_bucket(line, +/-3) + normalize(title)`. When fingerprints match, merge: keep highest severity, keep highest confidence, preserve the exact fingerprint, and retain the input IDs that produced the merged entry. A singleton that passes the gate is `surviving`; each input in a multi-input merge is provisionally `merged`.
 4. **Cross-reviewer agreement.** When 2+ independent reviewers flag the same issue (same fingerprint), boost the merged confidence by 0.10 (capped at 1.0). Cross-reviewer agreement is strong signal -- independent reviewers converging on the same issue is more reliable than any single reviewer's confidence. Preserve the distinction in the merged finding's artifact provenance according to the [synthesis artifact contract](./references/synthesis-artifact-contract.md).
@@ -480,7 +480,7 @@ Before applying the confidence gate, assign every finding in a valid return a st
 9. **Sort.** Order by severity (P0 first) -> confidence (descending) -> file path -> line number.
 10. **Collect coverage data.** Union residual_risks and testing_gaps across reviewers.
 11. **Preserve CE agent artifacts.** Keep the learnings, agent-native, schema-drift, and deployment-verification outputs alongside the merged finding set. Do not drop unstructured agent output just because it does not match the persona JSON schema.
-12. **Keep the input ledger complete.** Every enumerated input finding must receive exactly one final disposition and reason; reconcile the ledger according to the [synthesis artifact contract](./references/synthesis-artifact-contract.md).
+12. **Keep the input ledger complete.** Reconcile admitted findings and rejected-payload summaries according to the [synthesis artifact contract](./references/synthesis-artifact-contract.md).
 
 ### Stage 5b: Validation pass
 
@@ -519,7 +519,7 @@ Assemble the final report using **pipe-delimited markdown tables for findings**
 9. **Agent-Native Gaps.** Surface agent-native-reviewer results. Omit section if no gaps found.
 10. **Deployment Notes.** If deployment-verification-agent ran, surface the key Go/No-Go items: blocking pre-deploy checks, the most important verification queries, rollback caveats, and monitoring focus areas. Keep the checklist actionable rather than dropping it into Coverage.
 11. **Coverage.** Suppressed count, residual risks, testing gaps, failed/timed-out reviewers, validator failures, and any intent uncertainty carried by non-interactive modes.
-12. **Verdict.** Ready to merge / Ready with fixes / Not ready. Fix order if applicable. When an `explicit` plan has unaddressed requirements, the verdict must reflect it — a PR that's code-clean but missing planned requirements is "Not ready" unless the omission is intentional. When an `inferred` plan has unaddressed requirements, note it in the verdict reasoning but do not block on it alone.
+12. **Verdict.** Ready to merge / Ready with fixes / Not ready. Fix order if applicable. When an `explicit` plan has unaddressed requirements, the verdict must reflect it — a PR that's code-clean but missing planned requirements is "Not ready" unless the omission is intentional. When an `inferred` plan has unaddressed requirements, note it in the verdict reasoning but do not block on it alone. Apply the risk-aware degraded verdict rule from the [synthesis artifact contract](./references/synthesis-artifact-contract.md).
 
 Do not include time estimates.
 

package/skills/ce-review/references/synthesis-artifact-contract.md

@@ -19,11 +19,13 @@ reason. An unfinished `in_progress` artifact is evidence of an abnormal run,
 not evidence of a clean run. Never infer a clean run from an absent artifact.
 
 The artifact is parent-owned. Per-agent full-detail JSON files are written
-only after the persona return passes full-schema and environment-value
-validation. Rejected or never-returned personas do not produce per-agent
-files. If a later confidence or validation stage changes an input disposition,
-the parent updates the record and synthesis ledger before finalizing the
-artifact.
+only for findings admitted after the parent completes schema and
+environment-value validation. A finding rejected by environment-value
+detection is not persisted; other findings from the same return may proceed.
+A payload rejected at top level, or a rejected or never-returned persona,
+does not produce a per-agent file. If a later confidence or validation stage
+changes an input disposition, the parent updates the record and synthesis
+ledger before finalizing the artifact.
 
 ## Required distinctions and reconciliation
 
@@ -41,10 +43,15 @@ The artifact must preserve these distinctions:
       "dispatch_outcome": "findings",
       "input_finding_count": 2
     },
+    {
+      "persona": "testing",
+      "dispatch_outcome": "findings",
+      "input_finding_count": 1
+    },
     {
       "persona": "kieran-typescript",
       "dispatch_outcome": "malformed",
-      "input_finding_count": 1,
+      "input_finding_count": 2,
       "rejection_reason": "Rejected persona kieran-typescript return: field findings[0].evidence failed schema validation."
     }
   ],
@@ -55,6 +62,14 @@ The artifact must preserve these distinctions:
       "confidence": 0.55,
       "disposition": "suppressed",
       "reason": "confidence 0.55 is below the 0.60 gate"
+    },
+    {
+      "reviewer": "kieran-typescript",
+      "dispatch_outcome": "malformed",
+      "rejected_finding_count": 2,
+      "rejected_severities": ["P2", "P3"],
+      "disposition": "rejected",
+      "reason": "Rejected persona kieran-typescript return: field findings[0].evidence failed schema validation."
     }
   ],
   "findings": [
@@ -73,7 +88,7 @@ The artifact must preserve these distinctions:
     "merged": 2,
     "suppressed": 1,
     "filtered": 0,
-    "rejected": 0
+    "rejected": 2
   }
 }
 ```
@@ -84,16 +99,24 @@ The artifact must preserve these distinctions:
   naming persona and field without echoing the offending value. Dispatch
   outcome is separate from finding disposition.
 - `input_findings` is the authoritative parent-owned ledger. Before the
-  confidence gate, every safely enumerable finding receives an `input_id` of
-  `<reviewer>#<1-based finding index>`. Every enumerated input has exactly one
-  final `disposition`: `surviving`, `merged`, `suppressed`, `filtered`, or
-  `rejected`, plus a reason. Disposition counts equal the input-finding count.
-  If a rejected return has a safely enumerable `findings` array, assign IDs and
-  record each enumerated input as `rejected`.
-  A malformed JSON return with no safely enumerable finding has zero ledger
-  entries, not a fabricated finding. A rejected payload's reason is the exact
-  safe rejection message, not a bucket such as `invalid`; never include the
-  offending value.
+  confidence gate, every admitted finding receives an `input_id` of
+  `<reviewer>#<1-based finding index>`. Every admitted input has exactly one
+  final `disposition`: `surviving`, `merged`, `suppressed`, or `filtered`, plus
+  a reason. A rejected payload is represented by one summary ledger entry,
+  carrying the persona name, its `dispatch_outcome`, the
+  `rejected_finding_count` of findings not admitted, `disposition: "rejected"`,
+  `rejected_severities`, a list of the severities of the findings not
+  admitted as parsed from the payload, and the exact safe rejection message.
+  When a rejected finding's severity is absent, malformed, or not a valid
+  severity value, record it as `unknown`. Severity is metadata; recording it
+  never includes the offending value. Do not enumerate rejected findings or
+  assign them input IDs. A finding-level environment rejection uses the same
+  summary entry while admitted findings from that return continue normally.
+  Disposition counts are weighted by `rejected_finding_count` for that summary
+  entry, so their sum equals the total number of findings observed, not the
+  number of ledger rows. A malformed JSON return with no safely enumerable
+  finding has zero ledger entries, not a fabricated finding. Never include the
+  offending value in a rejection reason.
 - Synthesized and filtered findings retain their original fields plus
   `input_finding_ids` and provenance. Provenance contains the exact dedup
   fingerprint `normalize(file) + line_bucket(line, +/-3) + normalize(title)`,
@@ -120,8 +143,66 @@ the verdict is finalized. Existing artifacts without this additive metadata
 remain valid, with downstream consumers falling back to file mtime.
 
 Validation and persistence remain parent-side: no per-agent record or finding
-is written or merged until the complete persona payload passes schema and
-environment-value validation. Rejected or malformed persona returns do not
-fail the whole review; the review degrades while conforming returns continue
-through synthesis. Only an orchestration or storage failure that prevents the
-parent from producing the required run artifact is run-fatal.
+is written or merged until that finding passes schema and environment-value
+validation. Rejected findings are recorded through the single rejected-payload
+ledger summary; admitted findings from the same return remain eligible for
+synthesis. Rejected or malformed persona returns do not fail the whole review;
+the review degrades while conforming returns continue through synthesis. Only
+an orchestration or storage failure that prevents the parent from producing the
+required run artifact is run-fatal.
+
+## Environment-value validation
+
+The parent recursively inspects every string leaf without logging the raw
+return or any matched value. Structural environment detectors remain
+unbounded and unchanged: `$NAME`, `${NAME}`, `process.env.NAME`,
+`os.environ[...]`, and `NAME=value` assignments using a known environment
+variable name are shape-based checks.
+
+Value-based matching uses only non-empty runtime environment values that are
+at least 16 characters long and are not composed solely of digits, dots,
+dashes, or path-separator characters (forward slash or backslash). A
+value is also eligible regardless of length when
+its variable name contains one of `TOKEN`, `SECRET`, `KEY`, `PASSWORD`,
+`PASSWD`, `CREDENTIAL`, `AUTH`, `SESSION`, `COOKIE`, `PRIVATE`, `_PASS`,
+`_PWD`, `PASSPHRASE`, or `_SALT`, matched as a case-insensitive substring.
+Entries containing an underscore are matched against the variable name as
+written; the underscore is deliberate and prevents matching benign names that
+merely contain the bare word. Values that satisfy neither condition are not
+matched. A match is an exact or embedded match.
+
+If the offending string is inside one finding, drop that finding and record it
+through the rejected-payload summary entry; the remaining findings continue
+through validation and synthesis. If the offending string is outside any
+finding, reject the whole payload. Every rejection uses only the persona name,
+JSON path, and a fixed reason (`schema validation`, `environment-value
+detection`, or `malformed JSON`):
+`Rejected persona <name> return: field <JSON path> failed <reason>.` Never
+echo the matched value.
+
+## Risk-aware degraded verdict
+
+The risk-critical surfaces are `security`, `data-migrations`, `api-contract`,
+`reliability`, and `performance`. They are the conditional personas selected
+specifically for the matching diff shape in Stage 3. If one of those selected
+personas has `dispatch_outcome: "malformed"` or
+`dispatch_outcome: "never_returned"`, the review verdict must not be clean:
+it is blocking unless another persona covered the same surface and returned
+validated evidence for it. For this rule, validated evidence means at least
+one finding from that other persona's return passed complete schema and
+environment-value validation and is relevant to the same surface. A coverage
+note alone cannot satisfy this rule; the verdict must reflect the missing
+risk-critical evidence.
+
+Finding-level rejection is keyed by the severities in
+`rejected_severities`. A selected risk-critical persona whose rejected
+findings include any `P0`, `P1`, or `unknown` severity is treated exactly as a
+rejected persona for this verdict rule: blocking unless another persona
+covered the same surface with validated evidence. A selected risk-critical
+persona whose rejected findings are only `P2` or `P3` does not block on that
+basis alone; record it in the Coverage section instead. Unknown severity is
+treated as blocking as deliberate fail-closed behavior because the parent
+could not determine what was lost. Admitted findings and verdict blocking
+are independent: surviving findings from the same return continue through
+synthesis normally. Partial return is not partial coverage when the lost
+part was critical.