Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/not-diamond/self-care/helpgit clone --depth 1 https://github.com/Not-Diamond/self-careWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00010 | $0.01582 |
| Opus 5 | $0.00005 | $0.00791 |
| Sonnet 5 | $0.00002 | $0.00316 |
| Haiku 4.5 | $0.00001 | $0.00158 |
Grade A, and why
help scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 214 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Self-Care Help
Display help information about Self-Care commands, available detection skills, and usage examples.
Output
Display the following help text to the user:
╔══════════════════════════════════════════════════════════════════════════════╗
║ Self-Care — Help ║
╚══════════════════════════════════════════════════════════════════════════════╝
Self-Care is an agent-based trace analysis tool that validates agent execution
traces and detects quality issues using 14 specialized detection skills.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
COMMANDS
/self-care:init
Initialize Self-Care for this project.
Configures trace source (LangSmith, LangFuse, or local files),
validates credentials, and sets up the .self-care/ directory.
Required env vars (depending on source):
LANGSMITH_API_KEY For LangSmith
LANGFUSE_PUBLIC_KEY For LangFuse
LANGFUSE_SECRET_KEY For LangFuse
/self-care:analyze <trace-file> [--hash <hash>] [--skip-validation]
Validate and analyze a trace file for cases.
Outputs: Detection results in JSON format. Results are persisted per
trace file for tracking new vs. recurring issues.
Options:
--hash <hash> Pre-computed trace hash (16-char hex string).
Skips re-hashing and uses for memory lookup.
--skip-validation Skip format validation. Auto-detects format directly.
/self-care:run [trace-file] [details]
Run the full Self-Care pipeline on a trace.
If a trace file is provided, analyzes it directly.
If no file is provided, connects to the configured trace source
(from /self-care:init) and presents an interactive trace picker.
Orchestrates 6 sequential stages:
1. Validation — Verify trace format (OTEL or Claude Code)
2. Analysis — Detect cases using skill agents
3. Reporting — Generate triage report
4. Finding Review — Review each finding for relevance
5. Output Lists — Auto-remediable cases + cases needing context
6. Feedback — Optional anonymous feedback
Supports both OTEL (JSON) and Claude Code (JSONL) formats.
/self-care:review [count]
Review unreviewed reports from previous runs.
Finds reports with pending review status in .self-care/reports/
and walks through each finding for relevance classification.
Produces two categorized reports (auto-remediable + manual-review).
Optionally specify a count to limit how many reports to review
(newest first).
Examples:
/self-care:review — Review all pending reports
/self-care:review 3 — Review the 3 most recent pending reports
/self-care:autosync-enable
Enable scheduled autosync for the configured remote trace source.
Reuses .self-care/config.json, creates or updates
.self-care/autosync.json, and schedules /self-care:autosync-tick.
/self-care:autosync-disable
Disable scheduled autosync and cancel the scheduled tick task.
/self-care:autosync-status
Show current autosync settings, pending traces, and last sync state.
/self-care:autosync-tick
Run one non-interactive autosync cycle.
This is intended for Claude Code scheduled tasks and should not be run
manually unless you want to force one background-style pass.
/self-care:validate <trace-file>
Validate a trace file and extract metadata.
Detects trace format and reports:
• For OTEL: trace ID, service name, span count, time range
• For Claude Code: session ID, message count, event types, time range
/self-care:help
Display this help information.
/self-care:config [subcommand]
View or change Self-Care configuration settings.
Subcommands:
(no args) Show current configuration
analytics on|off Enable/disable usage analytics
disable <skill> Disable a detection skill
enable <skill> Re-enable a detection skill
severity <skill> <level> Override severity (high|medium|low)
severity <skill> reset Remove severity override
exclude <pattern> Add trace exclusion pattern
include <pattern> Remove trace exclusion pattern
autofix auto|prompt|disabled Set auto-fix behavior
reset Reset analysis config to defaults
Config is stored at: .self-care/config.json
/self-care:context [show|reset]
View or edit agent context for Self-Care analysis.
Describe your agent's expected behavior to help Self-Care
distinguish intentional patterns from problems.
Subcommands:
(no args) Prompt to edit agent context
show Display current agent context
reset Reset to default template
Context is stored at: .self-care/context.md
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
DETECTION SKILLS (14 case types)
Self-Care detects the following quality issues in agent execution traces:
• ambiguous-instructions — Unclear or conflicting instructions to agent
• context-utilization — Failure to use available context effectively
• contradictory-instructions — Instructions that conflict with each other
• goal-drift — Agent deviates from stated goal
• grounding — Loss of reference to base context/facts
• guardrail-violation — Breach of safety constraints
• instruction-following — Failure to follow explicit instructions
• missed-action — Task or action not performed when required
• missing-context — Insufficient context for task execution
• persona-adherence — Agent deviates from expected persona/role
• premature-termination — Task ends before completion
• reasoning-action-mismatch — Reasoning doesn't match taken actions
• step-repetition — Repeated steps without progress
• tool-failure — Tool execution failure or misuse
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
QUICK START
1. Set up a trace source:
/self-care:init
2. (Optional) Describe your agent's behavior:
/self-care:context
3. Enable background autosync:
/self-care:autosync-enable
4. Analyze a trace (interactive picker):
/self-care:run
5. Analyze a specific file:
/self-care:run ./trace.json
6. Review pending reports:
/self-care:review
7. Check autosync status:
/self-care:autosync-status
8. Validate trace format:
/self-care:validate ./trace.json
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
TRACE FORMATS
Self-Care supports:
• OTEL (OpenTelemetry) — JSON traces with span-based analysis
• Claude Code (JSONL) — Message-based traces from Claude Code sessions
Both formats are auto-detected. Use /self-care:validate to check format.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
For more information, visit the documentation at:
https://github.com/notdiamond/self-care
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 214 lines · 10 tokens per session scan A 9567fb3bc577
help is a command published in the GitHub repository Not-Diamond/self-care (28 stars, last pushed 4mo ago), licensed MIT. It adds 10 tokens to every session and 1,582 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
coder-eval-implement-plan
Implement an approved codereval plan phase by phase with risk-scaled per-phase review, then a final code review.
VibeGuard: Stats
View hooks trigger statistics - interception/warning/release times and reason analysis.
closeout
Close out a UNITARES workspace before final response.
checkin
Manual UNITARES governance check-in for the current turn.
re-review
Round 2+ review — confirm revisions landed, catch regressions, do not re-litigate.
project
Follow the project management workflow defined in .agents/skills/project/SKILL.md.