@fro.bot/systematic 3.34.0 → 3.35.0
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/package.json
CHANGED
package/skills/ce-plan/SKILL.md
CHANGED
|
@@ -349,6 +349,8 @@ For each question, decide whether it should be:
|
|
|
349
349
|
- **Resolved during planning** - the answer is knowable from repo context, documentation, or user choice
|
|
350
350
|
- **Deferred to implementation** - the answer depends on code changes, runtime behavior, or execution-time discovery
|
|
351
351
|
|
|
352
|
+
When you settle a question, or accept a requirement the plan builds on, and the answer rests on a claim that nothing in the origin document, the repo, or the research has checked (for example a requirement relayed from a meeting or a chat), and the answer affects architecture, scope, sequencing, or risk, load the `counterfactual` skill first if it is available. It comes from the separate metacogs plugin and is not a Systematic skill: load it by that bare name through the host's skill-loading mechanism, in the parent agent only, never in a dispatched sub-agent. Do not load it when the claim is already checked or the answer is mechanical. Carry its flip test into the plan as an assumption or open question rather than treating the claim as settled, with or without a user to ask. If the skill is not available, continue unchanged.
|
|
353
|
+
|
|
352
354
|
Ask the user only when the answer materially affects architecture, scope, sequencing, or risk and cannot be responsibly inferred. Use the platform's blocking question tool when available (see Interaction Method).
|
|
353
355
|
|
|
354
356
|
**Do not** run tests, build the app, or probe runtime behavior in this phase. The goal is a strong plan, not partial execution.
|
|
@@ -194,6 +194,8 @@ If agent outputs conflict:
|
|
|
194
194
|
- Prefer official framework documentation over secondary best-practice summaries when the conflict is about library behavior
|
|
195
195
|
- If a real tradeoff remains, record it explicitly in the plan
|
|
196
196
|
|
|
197
|
+
The metacogs skills `synthesis` and `counterfactual`, if you load them for such a conflict, come from a separate plugin and are not Systematic skills. Load them by their bare names through the host's skill-loading mechanism, in the parent agent only. If they are not available, continue unchanged.
|
|
198
|
+
|
|
197
199
|
## 5.3.6b Interactive Finding Review (Interactive Mode Only)
|
|
198
200
|
|
|
199
201
|
Skip this step in auto mode — proceed directly to 5.3.7.
|
|
@@ -603,6 +603,8 @@ The parent-owned artifact and its reconciliation rules are defined in the [synth
|
|
|
603
603
|
3. **Adjudicate every candidate group.** For each group, decide whether its members describe the same underlying defect (merge) or genuinely different defects (decline). Adjacency creates a candidate, not a conclusion -- findings on the same line describing different defects must stay declined. For a merge decision, write the merged finding's `title`, `why_it_matters`, `evidence`, and `line`, plus (when the route should narrow) a `proposed_route` with a `route_narrowing_reason`; optionally note `disagreement_facts` when reviewers disagreed on severity/autofix_class/owner, and `eligible_agreement_credit` for personas that agree without their own input finding in the group. For a decline decision, write a `declined_reason`. Every candidate-group member must be cited by exactly one decision -- no omissions, no double-citations.
|
|
604
604
|
|
|
605
605
|
Worked example: at `src/lib/model-availability.ts:139`, reliability's `Config hook awaits providers API without a timeout` and adversarial's `Config startup can hang forever behind a stalled /config` describe the same underlying defect in different words, so they merge.
|
|
606
|
+
|
|
607
|
+
If two reviewers in one group state opposed severities, or opposed claims about whether the defect is real or already handled, load the `synthesis` skill before deciding that group, if it is available and the mode is interactive or autofix (never headless or report-only). It comes from the separate metacogs plugin: it is not a Systematic skill and not this stage's merge. Load it by that bare name through the host's skill-loading mechanism, in the parent agent only, never in a reviewer or validator sub-agent. Rewording of one defect is not a disagreement. If the skill is not available, continue unchanged.
|
|
606
608
|
4. **Run `merge`.** It applies the adjudication envelope to `prepare`'s output, deriving each merged finding's severity, the cross-reviewer-agreement-boosted confidence (+0.10, capped at 1.0, for a merge with 2+ independent submitters), the fingerprint (`normalize(file) + "|" + line`), and the conservatively narrowed `autofix_class`/`owner`/`requires_verification` -- synthesis may narrow a route, never widen it without new evidence. It also returns `validator_requests`: the merged findings that need Stage 5b validation. See [pipeline invocation: merge](./references/pipeline-invocation.md#merge).
|
|
607
609
|
5. **Preserve CE agent artifacts.** Keep the outputs of the selected learnings, agent-native, schema-drift, and deployment-verification agents alongside the merged finding set for Stage 6 rendering. Do not drop unstructured agent output just because it does not match the persona JSON schema.
|
|
608
610
|
|
|
@@ -332,6 +332,8 @@ If agent outputs conflict:
|
|
|
332
332
|
- Prefer official framework documentation over secondary best-practice summaries when the conflict is about library behavior
|
|
333
333
|
- If a real tradeoff remains, record it explicitly in the plan rather than pretending the conflict does not exist
|
|
334
334
|
|
|
335
|
+
The metacogs skills `synthesis` and `counterfactual`, if you load them for such a conflict, come from a separate plugin and are not Systematic skills. Load them by their bare names through the host's skill-loading mechanism, in the parent agent only. If they are not available, continue unchanged.
|
|
336
|
+
|
|
335
337
|
### Phase 5: Synthesize and Rewrite the Plan
|
|
336
338
|
|
|
337
339
|
Strengthen only the selected sections. Keep the plan coherent and preserve its overall structure.
|