Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run a one-call pre-flight check on a single existing symbol before editing it: blast radius, co-change partners, complexity, 30-day churn, and verification checklist, then check Cortex decision memory for rationale/bans when intent may matter. Use before modifying any existing function or symbol you did not just write. Do not start editing a non-trivial existing function without pre-flight plus decision-memory recall/provenance; Memtrace knows the dependency graph, change history, and recorded d
.claude/skills/syncable-dev-memtrace-preflight/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 19% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 8% | 0% |
Pre-flight check: the two minutes before an edit that prevent the bug. One preflight_check call composes the dependency graph, the bi-temporal change history, and the complexity metrics into an actionable radar for a single symbol.
| Tool | Purpose | |------|---------| | preflight_check | Full radar for one symbol: blast radius + co-change + churn + checklist | | get_impact | Deeper blast-radius walk when the radar flags HIGH/CRITICAL | | get_cochange_context | Wider co-change window when partners look surprising | | recall_decision | Recorded choices, bans, and conventions for the symbol/subsystem/behavior | | why_is_this_here / governing_contracts | Symbol-scoped rationale and constraints when a symbol id is available |
> Parameter types: MCP parameters are strictly typed. Numbers must be
Full parameter spec for every Memtrace tool: references/mcp-parameters.md (bundled at the memtrace-skills plugin root). > JSON numbers — not strings.
preflight_check with:
repo_id — requiredsymbol — the function/method name you are about to modifyEach line names something concrete to verify:
get_impact again after your edit andcompare total_affected; new dependents mean your change leaked wider than intended.
hot code punishes large diffs.
correctly after the change.
assuming the current shape is intentional.
Files that historically changed together with the target usually need updating in the same change. If you finish the edit without touching a frequent partner, say so explicitly and why.
preflight_check (or get_impact) on the symbol.if it did, simplify before declaring the work done.
Before changing behavior, deleting code, or refactoring odd/legacy code:
recall_decision("<symbol/subsystem/behavior>").symbol_id, runwhy_is_this_here(symbol_id) and governing_contracts(symbol_id).
risk: CRITICAL/HIGH — propose the change plan before editing; considera feature flag or staged rollout.
dependents: 0 — edit freely; note the symbol may be newly added or anentry point.
contradict it without explicit user sign-off.
preflight_check returns one radar for the symbol:
| Field | Meaning | |-------|---------| | risk | verdict, e.g. HIGH / CRITICAL (see Verdict guidance) | | dependents | blast radius — count of symbols that depend on the target | | co-change partners | files that historically change with it (step 3) | | complexity + 30-day churn | current complexity score and recent change count | | checklist | generated verification lines, e.g. "hotspot: 6 changes in 30 days", "sits in 2 process flow(s)" |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-15 | fail→pass | 10,344 | 2,758 | -73% | 1 | 1 | 0% | 1,460 | 1,332 | -9% | 0 | 0 | — |
case-01 | fail→fail | 21,780 | 5,630 | -74% | 1 | 1 | 0% | 3,362 | 1,335 | -60% | 0 | 0 | — |
case-02 | fail→fail | 17,110 | 6,316 | -63% | 1 | 1 | 0% | 2,642 | 1,248 | -53% | 0 | 0 | — |
case-03 | fail→fail | 16,054 | 6,294 | -61% | 1 | 1 | 0% | 2,680 | 1,295 | -52% | 0 | 0 | — |
case-04 | pass→fail | 10,829 | 14,023 | +29% | 1 | 1 | 0% | 1,648 | 2,152 | +31% | 0 | 0 | — |
case-05 | pass→pass | 17,196 | 40,865 | +138% | 1 | 1 | 0% | 3,511 | 5,469 | +56% | 0 | 0 | — |
case-06 | pass→pass | 4,881 | 6,428 | +32% | 1 | 1 | 0% | 940 | 2,050 | +118% | 0 | 0 | — |
case-07 | fail→pass | 12,156 | 6,156 | -49% | 1 | 1 | 0% | 1,962 | 1,758 | -10% | 0 | 0 | — |
case-08 | pass→pass | 12,866 | 8,857 | -31% | 1 | 1 | 0% | 1,993 | 2,402 | +21% | 0 | 0 | — |
case-09 | fail→pass | 12,040 | 7,592 | -37% | 1 | 1 | 0% | 1,811 | 2,147 | +19% | 0 | 0 | — |
case-10 | pass→pass | 5,951 | 3,401 | -43% | 1 | 1 | 0% | 999 | 1,499 | +50% | 0 | 0 | — |
case-11 | pass→pass | 8,606 | 4,534 | -47% | 1 | 1 | 0% | 1,359 | 1,632 | +20% | 0 | 0 | — |
case-12 | fail→pass | 7,311 | 4,619 | -37% | 1 | 1 | 0% | 1,204 | 1,713 | +42% | 0 | 0 | — |
case-13 | fail→pass | 8,119 | 4,144 | -49% | 1 | 1 | 0% | 1,297 | 1,405 | +8% | 0 | 0 | — |
case-14 | pass→pass | 6,197 | 1,946 | -69% | 1 | 1 | 0% | 1,020 | 1,200 | +18% | 0 | 0 | — |
case-16 | fail→pass | 19,530 | 6,806 | -65% | 1 | 1 | 0% | 752 | 2,071 | +175% | 0 | 0 | — |
case-17 | fail→pass | 13,806 | 13,216 | -4% | 1 | 1 | 0% | 1,915 | 1,632 | -15% | 0 | 0 | — |
case-18 | pass→pass | 11,609 | 6,367 | -45% | 1 | 1 | 0% | 1,831 | 1,978 | +8% | 0 | 0 | — |
case-19 | pass→pass | 4,329 | 2,157 | -50% | 1 | 1 | 0% | 685 | 1,238 | +81% | 0 | 0 | — |
case-20 | fail→pass | 6,103 | 3,058 | -50% | 1 | 1 | 0% | 880 | 1,254 | +43% | 0 | 0 | — |
case-21 | fail→pass | 4,823 | 3,460 | -28% | 1 | 1 | 0% | 695 | 1,449 | +108% | 0 | 0 | — |
case-22 | fail→pass | 9,549 | 3,817 | -60% | 1 | 1 | 0% | 1,373 | 1,579 | +15% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 17 counted toward the lift figure. The other 5 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +41 percentage points is the difference between those two pass rates over the 17 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.