Elgato_dark/ring
1
0
Fork 0
mirror of https://github.com/LerianStudio/ring synced 2026-04-21 13:37:27 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

docs(shared-patterns): add anti-rationalization table to gate-cadence-classification

Browse source 
CLAUDE.md mandates anti-rationalization tables for files read by agents as authoritative references. gate-cadence-classification.md is high-risk for rationalization because AI under time pressure tends to reason 'running a task-cadence gate per subtask is safer — more runs'. The new table defends against four common drifts: per-subtask inflation of task-cadence gates, shifting cycle-cadence work to task end, bypassing cadence for single-subtask tasks, and using cycle brevity as an excuse to skip classification.

X-Lerian-Ref: 0x1
This commit is contained in:
Fred Amaral 2026-04-18 10:44:53 -03:00
parent 5926ef7515
commit f760f24131
No known key found for this signature in database
GPG key ID: ADFE56C96F4AC56A
1 changed files with 9 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
dev-team/skills/shared-patterns/gate-cadence-classification.md
View file

@ -44,3 +44,12 @@ Sub-skills that run at task cadence MUST accept aggregated input:
Sub-skills that run at subtask cadence MUST continue to accept scoped input:
- `implementation_files`: array (this subtask's changes only)
- `gate0_handoff`: object (this subtask's handoff)
## Anti-Rationalization Table
| Rationalization | Why It's WRONG | Required Action |
|-----------------|----------------|-----------------|
| "Running a task-cadence gate per subtask is safer — more runs catch more bugs" | Task-cadence gates (devops, SRE, fuzz, property, integration-write, chaos-write, review) operate on the UNION of subtask outputs. Per-subtask firing wastes cycle time on in-flight code that will be re-tested when the task completes. Fuzz seeds and property shrinkers stabilize at task boundary; earlier firing produces noisier results. | **MUST dispatch task-cadence gates once per task, after ALL subtasks have passed their subtask-cadence gates (0, 3, 9).** |
| "A cycle-cadence gate can run at task end — close enough" | Cycle-cadence gates (integration-execute, chaos-execute, multi-tenant-verify, migration-safety, dev-report) require shared infrastructure (containers, Toxiproxy, tenant fixtures) whose setup/teardown cost dominates the actual check. Firing them per task inflates cycle duration by the number of tasks. | **MUST defer execute-mode infrastructure gates to cycle end (Step 12.x of dev-cycle).** |
| "This task has only one subtask, so cadence doesn't matter" | Cadence is a schema-level invariant enforced by `validate-gate-progression.sh` and the state-write paths documented in `dev-cycle/SKILL.md`. Bypassing it writes state to the wrong path and breaks the hook's progression check for the next task that has multiple subtasks. | **MUST follow the documented cadence regardless of subtask count. Treat single-subtask tasks as "subtasks: [task-itself]" for state purposes.** |
| "I'll run all gates per subtask because the cycle is short anyway" | Cycle brevity does not license cadence violation. The cadence model is also how reviewers consume aggregate context; per-subtask firing produces incomplete review inputs. | **MUST classify each gate against this table before dispatch. When unclear, STOP and ask the orchestrator.** |