Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 22 additions & 0 deletions claude/skills/retrospective/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,7 @@ recurrence through external mechanisms:
- Script or automation
- Template or checklist
- Permission setting
- CI check (GitHub Actions workflow, test assertion, etc.)
- Project structure change

Format as a numbered list of pairs. Include a severity rating for each:
Expand All @@ -41,6 +42,24 @@ Severity: high / medium / low (with brief justification: time wasted, user inter
Countermeasure: <structural fix> (scope: global / project-specific)
```

### Countermeasure Specificity

Each countermeasure must specify:

- The exact target file path (e.g., `AIRULES.md`, `CLAUDE.md`,
`.claude/rules/foo.md`, `.pre-commit-config.yaml`)
- The concrete content to add or modify — draft the actual wording, not
just "add a rule about X"

### Placement Scope

Choose the placement based on who benefits:

- Team-shared enforcement → project's `CLAUDE.md` or its `rules/` directory
- Personal habits or preferences → global `AIRULES.md` or `~/.claude/rules/`
- Memory is for recording facts only. It must not be used as a mechanism
for behavior change (see Prohibited Countermeasures below).

## Prohibited Countermeasures

The following are NOT valid countermeasures. Never propose them:
Expand All @@ -51,6 +70,9 @@ The following are NOT valid countermeasures. Never propose them:
- "Remember to Z"
- Any countermeasure that relies on the model's judgment, attention, or
memory improving in the future
- "Save to memory" / "Record in memory" — memory depends on model
judgment for recall and is not a structural enforcement mechanism.
Memory is valid for recording facts, but not for driving behavior change.
- Restating an existing rule as if adding it would help

The premise: the model's behavior cannot be trusted to improve through
Expand Down