dont-forget
Skill0 starsChanged 2 months agoMITView on GitHub
Turn recurring obligations into executable automation (CI, hooks, scripts, codegen, lint) instead of prose reminders agents may skip. Use when the user asks to always, never forget, every time, on each deploy, before commit, or when a task was forgotten repeatedly and needs hard enforcement.
- Reads credentials
In this repository26 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .cursor/rules/automate-before-manual.mdcCursor rule
- .cursor/rules/finish-task-dev-server.mdcCursor rule
- brainstormingSkill
- caveman-commitSkill
- find-skillsSkill
- semantic-versionSkill
- systematic-debuggingSkill
- verification-before-completionSkill
- writing-plansSkill
- ask-before-architectureSkill
- automate-before-manualSkill
- caveman-commitSkill
- diagrams-mermaidSkill
- dont-be-lazySkill
- find-jobSkill
- finish-with-dev-serverSkill
- google-calendar-mcpSkill
- init-fmzSkill
- instagram-chrome-postSkill
- project-initSkill
- recapSkill
- semantic-versionSkill
- skills-sh-maintainerSkill
- ui-change-e2eSkill
- validate-before-shareSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 7 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 15
Default: when the user asks for something that must happen repeatedly or was forgotten before, encode it in executable machinery the next …
- Distinctive: line 64
Wait for user approval or adjustment: Do not implement until the user confirms.
- Distinctive: line 70
README.md: when operators without agents need it
- Distinctive: line 71
Never a paragraph asking future agents to remember
- Very specific: line 147
Add scripts/increment-deploy-count.sh (or inline in workflow).
- Very specific: line 148
Call it in .github/workflows/deploy.yml after successful deploy.
- Very specific: line 149
Commit deploy-count.json (or expose via API).
- Distinctive: line 152
Don't: Add only a user rule saying "remember to increment deploy count."
- Distinctive: line 156
After implementation (post-approval confirmation — not a substitute for the pre-implementation proposal), state:
- Distinctive: line 160
What fails: if skipped
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Don't forget — obligation as automation
Agents forget. Written rules and chat promises are soft — they compete with context, get skipped under pressure, and vanish in new sessions.
Default: when the user asks for something that must happen repeatedly or was forgotten before, encode it in executable machinery the next agent cannot bypass without breaking the build.
The agent picks the highest tier that fits this repo — CI step, hook, failing test, codegen diff, npm script chain, etc. The "software" is whatever executable gate matches the obligation and trigger; not a fixed recipe.
Not "AI safety guardrails" — enforcement: repo machinery that fails when an obligation was skipped.
When to use
- User says: always, every time, never forget, on each deploy, before every commit, don't let this slip again
- Same instruction was given in a prior session and not followed
- A counter, check, sync step, or enforcement must survive session handoff
- User prefers "put it in the build" over "remember to do X"
When NOT to use
- One-off task in the current session only
- Judgment-heavy work (copy tone, architecture tradeoffs) with no objective pass/fail
- Automation cost exceeds value (5-minute yearly task)
- User explicitly wants a manual checklist only
Enforcement ladder (prefer top)
| Tier | Mechanism | Survives new chat? | Example |
| 1 | CI / deploy pipeline | Yes | Bump deploy counter in GitHub Actions |
| 2 | Git hooks (pre-commit, pre-push) | Yes | Block commit if DER not updated |
| 3 | Build / test failure | Yes | make der-check in CI |
| 4 | Lint / type rules | Yes | ESLint rule, custom Ruff check |
| 5 | Codegen / scaffolder | Yes | npm run generate:api writes client from OpenAPI |
| 6 | Scheduled job | Yes | Cron syncs inventory nightly |
| 7 | Cursor hook / project rule | Partial | .cursor/hooks.json runs validator on save |
| 8 | Makefile / npm script | Partial | npm run deploy chains all required steps |
| 9 | Prose rule (AGENTS.md, SKILL.md) | Weak | Last resort or supplement to tiers 1–8 |
Pick the highest tier that fits. Add a prose rule only to point at the automation, not replace it.
Workflow
- Name the obligation — one sentence, observable outcome ("deploy counter increments on every production deploy").
- Find the natural trigger — deploy, commit, push, PR merge, file save, schedule, API call.
Choose enforcement tier from the ladder above. 3b. Propose to user (before any file edits) — Present in plain language:
- The obligation (one sentence, observable)
- The trigger (deploy, commit, PR, schedule, etc.)
- Chosen mechanism and tier (e.g. "GitHub Actions step after deploy — tier 1")
- Files to create or change
- What fails or blocks if the obligation is skipped
- Optional: one-line
AGENTS.mdpointer you will add after implementation
Wait for user approval or adjustment. Do not implement until the user confirms.
- After user approval, implement the smallest hard gate — script that exits non-zero, generated file diff, or pipeline step that fails.
- Wire the trigger — hook, workflow,
package.jsonscript chain, Makefiledeploytarget. - Verify — run the trigger locally; confirm failure when obligation is skipped and success when met.
- Document briefly — one line pointing at the automation:
- AGENTS.md (preferred in Cursor/agent repos): e.g.
Deploy counter: auto-incremented in .github/workflows/deploy.yml - README.md when operators without agents need it
- Never a paragraph asking future agents to remember
- AGENTS.md (preferred in Cursor/agent repos): e.g.
Decision checklist
Before finishing, confirm:
- Obligation is tied to a concrete event (not "sometimes")
- Skipping it causes visible failure (red CI, blocked commit, test fail) OR automatic correction (codegen)
- A new agent can discover it from repo files, not chat history
- User saw and approved the enforcement proposal before code changed
- User was not given "I'll remember" without code
Patterns by domain
Deploy / release
- GitHub Actions step after successful deploy
- Release script that bumps version + changelog + counter in one command
- Tag hook that refuses tag if checklist script fails
Commits / PRs
- pre-commit: format, secrets scan, schema ↔ DER sync
- PR CI: required check that cannot be skipped
- Danger / custom bot comment is weak — prefer failing check
Schema / docs drift
- Codegen from single source of truth (OpenAPI → client, SQL → types)
make checkcompares generated output to committed files- Fail CI if
database/schema.sqlchanged butdatabase/der.mddid not
Environment / ports / health
docker-composehealthcheck +depends_on: condition: service_healthy- Startup script that probes
/healthbefore reporting URL to user .env.examplevalidated by script in CI
Agent-specific enforcement
- Cursor hook on
afterFileEditrunning project validator .cursor/rulesthat say runnpm run verify-x, not "remember to verify"- Project skill with
disable-model-invocation: falseonly when prose truly helps
Recurring maintenance
cron/ GitHub scheduled workflow- Dependabot + auto-merge with passing CI
- Stale-issue bot is notification-only — prefer workflow that does the sync
Anti-patterns
| Bad | Better |
| "I'll increment the counter next deploy" | Step in deploy.yml increments deploy-count.txt |
| Long AGENTS.md section repeating the same reminder | pre-commit + one-line pointer |
| Optional npm script nobody runs | deploy script calls it; CI runs deploy --dry-run |
Comment // TODO: don't forget X | Test or lint that fails until X exists |
Example (deploy counter)
User: "Increment a counter every deploy — you forgot three times."
Propose (before coding):
Obligation:
deploy-count.jsonincrements on every successful production deploy. Trigger: GitHub Actions deploy workflow, after health check passes. Mechanism: Tier 1 — workflow step runsscripts/increment-deploy-count.shand commits or updates the counter file. Files:.github/workflows/deploy.yml,scripts/increment-deploy-count.sh,deploy-count.jsonIf skipped: deploy job fails (or counter stays stale — prefer failing the step). AGENTS.md line:Deploy count: see .github/workflows/deploy.yml
Approve to implement?
Do:
- Add
scripts/increment-deploy-count.sh(or inline in workflow). - Call it in
.github/workflows/deploy.ymlafter successful deploy. - Commit
deploy-count.json(or expose via API). - AGENTS.md:
Deploy count: see .github/workflows/deploy.yml
Don't: Add only a user rule saying "remember to increment deploy count."
Report to user
After implementation (post-approval confirmation — not a substitute for the pre-implementation proposal), state:
- What is enforced
- Where (file + trigger)
- What fails if skipped
- That future sessions do not rely on memory
Additional examples
See reference/examples.md for more scenarios and tier choices.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

