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/alexmond/alexmskills/helpgit clone --depth 1 https://github.com/alexmond/alexmskillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/alexmond/alexmskills/help)<a href="https://agentmods.dev/commands/alexmond/alexmskills/help"><img src="https://agentmods.dev/badge/commands/alexmond/alexmskills/help.svg" alt="Measured on agentmods" height="20"></a>What 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 | $0.00015 | $0.02613 |
| Opus 5 | $0.00008 | $0.01307 |
| Sonnet 5 | $0.00003 | $0.00523 |
| Haiku 4.5 | $0.00002 | $0.00261 |
Grade B, and why
help scanned grade B with 1 finding 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 3d 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.
Asks the agent to reveal its instructionsmediumSystem prompt leakage
Directions to print, repeat or translate the system prompt extract configuration the operator did not intend to expose.
description: Show prompt-coach help — commands, config options, and say-it phrases How it starts
The opening of the file, as written. The whole thing — 200 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/prompt-coach:help
Print a compact help card for the coach. Read the current installed version and resolved config to fill in the live values.
What to do
- Read
~/.claude/plugins/cache/alexmskills/prompt-coach/*/.claude-plugin/plugin.json(whichever version exists) for.versionand.description. - Resolve current effective config by merging (in order): built-in defaults →
~/.claude/prompt-coach/config.json→ current repo's.claude/prompt-coach/config.json(if any). - Render the card below with the live values substituted.
The card
Render as a fenced block so it presents like a dashboard, under 60 lines.
prompt-coach v<VERSION>
A UserPromptSubmit hook that watches every prompt you send Claude Code and nudges
you toward better prompting habits. 42 rules across 6 tiers, 42 positive detectors,
typo tolerance, conversational + picker-answer short-circuit. Rules quietly
graduate as you master them and fade to occasional refreshers.
New to the coach? See SKILL.md's "Quick start" section — a 60-second setup,
what a nudge looks like, the slash commands, and the most-used say-it
phrases. Path:
~/.claude/plugins/cache/alexmskills/prompt-coach/*/skills/prompt-coach/SKILL.md
────────────────────────────────────────────────────────────────────
COMMANDS
/prompt-coach:stats Health dashboard — prompts analyzed, emit rate,
top-fired rules, mastery, typo corrections, config
/prompt-coach:mastery Which rules mastered, which need reset — with
analysis (well-tested / barely-tested / untested);
surfaces close-to-mastery rules (v0.26+)
/prompt-coach:analyze On-demand analysis (v0.37+): run the FULL 42-rule
catalog against a pasted prompt or your last N
logged prompts, with a coached rewrite / pattern report
/prompt-coach:config Structured config surface — show / get / describe /
set / reset / diff / export / mastery / acceptance /
sources / dashboard (v0.18+)
/prompt-coach:library Match your task to the closest gold-standard template from
Anthropic's Claude Code Prompt Library (v0.47; offline snapshot)
/prompt-coach:dashboard Local web dashboard (v0.44) — stats, mastery by tier with
reference URLs, a live config editor, and a Library tab. Zero-
dependency stdlib server on 127.0.0.1. (Raw JSON: `config dashboard`.)
(Cross-repo daily review moved out in v0.23. Say "log review" or
"daily review" to invoke ~/.claude/skills/log-review/ instead.)
/prompt-coach:report-issue File a privacy-safe bug report for prompts the coach
mistreated. Uses candidates.jsonl seeded by the
say-it phrases below.
/prompt-coach:help This card.
────────────────────────────────────────────────────────────────────
SAY-IT PHRASES (talk to Claude naturally; Claude edits the right file)
On/off (v0.29+ — rendering is always inline):
"disable prompt-coach" set enabled: false in this scope
"enable prompt-coach" set enabled: true in this scope
"coach pause 10" temporary silence for N prompts
"reset prompt-coach mode" drop local override (legacy — no-op post-v0.29)
Analyze / sources / files (v0.36+):
"analyze this prompt: <text>" full-catalog read + a coached rewrite
"analyze my last N prompts" pattern report over recent history
"open the docs for <rule>" open the Anthropic guide URL in a browser
"show me the skill folders" list plugin folders, state, runnable scripts
Pause / disable:
"coach pause 10" silence for the next N prompts
"coach off <rule-id>" permanently disable one rule
"coach on <rule-id>" re-enable it
"coach reactivate <rule-id>" per-repo — re-open a rule that mastered globally
"disable praise" silence encouragement, keep nudges
"praise every N prompts" set praise_ratio
Flag a bad call (adds to candidates.jsonl for /report-issue):
"coach that was wrong" flags the PRIOR prompt for review
"coach missed this"
"coach false positive"
"bad nudge" / "wrong nudge"
────────────────────────────────────────────────────────────────────
CURRENT CONFIG (resolved: default → global → this repo)
enabled: <live value> (master switch; false = fully off)
ack_clean: <live value> (✓ heartbeat on clean prompts)
ack_ratio: <live value> (1 = every clean prompt)
show_source_urls: <live value> (clickable doc URLs in the coach block)
collaborator_gate: <live value> (false = honest proceed + name prompt; true = stop & wait)
min_demonstrations: <live value> (times you USED the technique → mastered, v0.40)
regression_guard: <live value> (clean prompts since last fire, needed w/ demos)
inactive_after: <live value> (clean_streak w/ 0 demos → inactive)
graduation_threshold: <live value> (clean-streak recency signal; not the mastery driver)
cooldown_prompts: <live value> (min between same-rule fires)
mastered_cooldown_prompts: <live value> (min between refresher fires)
max_active_rules: <live value> (default 6)
praise_ratio: <live value> (1 praise per N clean prompts w/ positive)
praise_on_mastery: <live value>
praise_on_first_after_fire: <live value>
disable_praise: <live value>
tips_enabled: <live value> (proactive advanced-technique tips)
typo_tolerance: <live value> (0 disables normalization)
disabled_rules: <live value>
demote_on_regression: <live value> (self-healing; off by default)
Global config file: ~/.claude/prompt-coach/config.json
Per-repo config file: .claude/prompt-coach/config.json (per repo)
────────────────────────────────────────────────────────────────────
CONFIG OPTIONS REFERENCE
enabled Master switch. false = hook returns immediately (default: true)
ack_clean ✓ liveness heartbeat on clean prompts (default: true)
ack_ratio Emit the ack every Nth clean prompt (default: 1)
show_source_urls Full clickable doc URLs in the coach block (default: true)
library_hints Ground the rewrite in the closest Prompt Library template (default: true)
collaborator_gate Make the collaborator block a real confirmation gate (default: false).
false = honest: block says it's proceeding, Claude answers the same
turn and states which prompt it used (▸ Working from …). true = Claude
STOPS after the block and waits for yes/no/edit (one extra round-trip).
min_demonstrations EARNED MASTERY (v0.40): times the mirroring positive must fire
— i.e. times you USED the technique — before a rule masters (default: 3)
regression_guard Clean prompts since last fire required alongside demos (default: 3)
inactive_after clean_streak with 0 demonstrations → rule retires inactive (default: 15)
graduation_threshold Clean-streak recency/decay signal; no longer drives mastery (default: 15)
min_fires_for_mastery Legacy v0.27 gate, superseded by min_demonstrations; ignored
ADAPTIVE COACHING (v0.41)
precision_gating Demote rules you consistently reject to dormant (default: true)
precision_floor Acceptance-rate floor below which a rule is demoted (default: 0.15)
min_outcomes_for_gating Recorded accept/edit/reject outcomes needed before gating (default: 4)
record_silence_as_accept Honest mode: moving on w/o objecting to a rendered rewrite = implicit accept (default: true)
explore_period Re-admit one dormant rule every N prompts (default: 10; 0=off)
max_nudges_per_window Max visible rewrites per nudge_window prompts (default: 6; 0=off)
nudge_window Rolling window (prompts) for the fatigue cap (default: 20)
review_intervals_days Decaying-mastery review schedule, days of non-use (default: 30/90/180)
cooldown_prompts Min prompts between same-rule nudges (default: 5)
mastered_cooldown_prompts Refresher cooldown for mastered rules (default: 50; 0=off)
max_active_rules Cap on practicing rules active at once (default: 6)
pause_until_prompt Skip nudging until global prompt_count > this
disabled_rules Array of rule ids to permanently silence
praise_ratio 1 praise per N clean prompts w/ positive fire (default: 10)
praise_on_mastery Celebrate rule mastery events (default: true)
praise_on_first_after_fire Praise the immediate correction (default: true)
disable_praise Silence encouragement, keep nudges (default: false)
praise_novelty_window Don't repeat a praise variant within N fires (default: 5)
TIPS (proactive advanced-technique nudges)
tips_enabled Emit occasional 💡 technique tips on clean prompts (default: true)
tip_ratio One tip per N eligible clean prompts (default: 5)
tip_cooldown_prompts Min prompts between tips (default: 100)
typo_tolerance Levenshtein edit-distance for typo normalization (default: 2)
demote_on_regression Object {enabled, threshold, window} — auto-demote mastered
rules that fire threshold+ times within window prompts.
Off by default.
llm_fallback Object {enabled, model, min_words} — opt-in stub for a future
LLM classifier when regex misses. Disabled by default.
(v0.38 removed the legacy `nudge_style`, `voice_preset`/`voice_source`, and the
anti-habituation keys along with the hand-written nudge path. The coach is
collaborator-only: when a rule fires, Claude rewrites your prompt fresh, so
there's no preset to pick and no repeated text to habituate to.)
────────────────────────────────────────────────────────────────────
STATE FILES
Global mastery ledger: ~/.claude/prompt-coach/state.json
Per-repo state: .claude/prompt-coach/state.json (this repo)
Fire log (audit): .claude/prompt-coach/log.md
Bug-report candidates: .claude/prompt-coach/candidates.jsonl
────────────────────────────────────────────────────────────────────
DOCS
Full SKILL.md: <plugin dir>/skills/prompt-coach/SKILL.md
Sources: <plugin dir>/docs/sources.md — bibliography per rule
Marketplace: alexmond/alexmskills → prompt-coach
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.
- 3d ago First seen · 200 lines · 15 tokens per session scan B e6b21ddb288a
help is a command published in the GitHub repository alexmond/alexmskills (6 stars, last pushed 3d ago), licensed MIT. It adds 15 tokens to every session and 2,613 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it B with 1 finding (asks the agent to reveal its instructions). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
git
Git operations with intelligent commit messages and workflow optimization.
checklist
Generate a custom checklist for the current feature based on user requirements.
clarify
Identify underspecified areas in the current feature spec by asking up to 5 highly targeted clarification questions and encoding answers back into the spec.
specify
Create or update the feature specification from a natural language feature description.
analyze
Perform a non-destructive cross-artifact consistency and quality analysis across spec.md, plan.md, and tasks.md after task generation.
converge
Assess the current codebase against the feature's spec, plan, and tasks, then append any remaining unbuilt work as new tasks to tasks.md so implement can complete it.