Phantom is an AI co-worker that operates its own computer, maintains persistent memory, uses MCP tools, and builds software and infrastructure in its workspace. It is for people who want an agent that can continue work across sessions through Slack or web chat while keeping its workspace separate from the user's computer. The catalogue skills and instruction support workflows for Phantom.
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 skills add ghostwright/phantom --skill overheardgit clone --depth 1 https://github.com/ghostwright/phantomWrote 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/skills/ghostwright/phantom/overheard)<a href="https://agentmods.dev/skills/ghostwright/phantom/overheard"><img src="https://agentmods.dev/badge/skills/ghostwright/phantom/overheard/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/ghostwright/phantom/overheard"><img src="https://agentmods.dev/badge/skills/ghostwright/phantom/overheard.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Anti-Refusal · line 80 Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.Fix: Remove instructions that suppress warnings, disclaimers, or ethical commentary. Let the agent surface safety-relevant caveats to the user.
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 5.1 | $0.00024 | $0.01138 |
| Opus 5.5 | $0.00010 | $0.00455 |
| Sonnet 5.5 | $0.00005 | $0.00228 |
| Haiku 4.5 | $0.00002 | $0.00114 |
Grade A, and why
overheard scanned grade A 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 1mo 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.
Strips warnings and disclaimerslowAnti-refusal
Omitting safety caveats hides risk from the user and is a common jailbreak preamble.
- Never moralize about broken commitments. The tone is "here is what I saw", not "here is what you should do". Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
How it starts
The opening of the file, as written. The whole thing — 86 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Overheard: the promises audit
Goal
Find promises the user made in conversations (in Slack, in email, in messages) that start with phrases like "I'll send", "I will", "let me get back to you", "by Friday", "next week", "I'll follow up", and check whether there is evidence in subsequent memory of follow-through. Surface the gaps as a promises audit, with context for each and a draft action.
The shadow backlog most people carry. The user does not need to be reminded of every forgotten ping; they need the two or three that actually matter.
Steps
1. Pull the last 14 days of memory
Call mcp__phantom-reflective__phantom_memory_search with query: "I will", days_back: 14, limit: 30. Then call it again with query: "let me" and merge. Then query: "follow up", and merge. Then query: "by Friday" and merge. The merge should deduplicate by episode id.
Success criteria: you have a pool of recent episodes that contain commitment language.
2. Extract candidate commitments
For each episode, read the detail and pull out phrases where the user was the one making the commitment, not receiving one. Common patterns:
- "I will send you..."
- "Let me get back to you on..."
- "I'll check with and let you know."
- "By Friday I'll have..."
- "Next week I'll..."
- "I'll follow up on..."
Skip commitments that are clearly to yourself, to the agent, or to no one in particular ("I will make more coffee"). Skip commitments that the user has already negated ("I thought I'd send X but actually I'm not going to").
Success criteria: you have a list of 5-15 candidate commitments with the original context, who the promise was to (if known), and when it was made.
3. Check for follow-through
For each candidate, search memory again for follow-up evidence. The query is a short paraphrase of what was promised. Example: if the commitment was "I'll send Anna the revised doc by Friday", search for "send Anna doc" or "revised doc Anna" with days_back: 14.
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.
- 1mo ago First seen · 86 lines · 24 tokens per session scan A 648928cf3e5c
overheard is a skill published in the GitHub repository ghostwright/phantom (1,484 stars, last pushed 4mo ago), licensed Apache-2.0. It adds 24 tokens to every session and 1,138 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 1 finding (strips warnings and disclaimers). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
scout-consolidation
Launch a Scout consolidation session as a background process. Runs a lighter delta scan — captures what changed since the last run and updates action items and KB.
handoff
Generate a HANDOFF.md that captures goal, current state, files touched, failed attempts, and next steps — so a fresh Claude session can continue exactly where this one left off. Use when sessions are getting long, Claude keeps retrying the same broken solution, or the user wants to step away and resume later.
hive.context-preservation
Proactively extract critical values from tool results into working notes before automatic context pruning destroys them.
hive.note-taking
Maintain a free-form scratchpad of decisions, extracted values, and open questions so context pruning doesn't lose anything you still need.
daily-briefing
Proactive daily briefing that fires on a recurring schedule, pulls recent memory and workspace context, composes a structured summary (action items, progress, radar, next steps), and delivers it to all active channels. Enable with a time like "set up my daily briefing at 9am". Disable, reschedule, or check status at…
identity-snapshot
Generate a living profile of the user's working patterns, decision tendencies and quality preferences from their Dex data. Use when the user says 'what are my patterns', 'how do I work', or during week-review. Also use proactively when the model (older than 7 days) is stale. Not for planning a week; use week-plan.