Eight skills for product marketers who need to turn competitor evidence into useful decisions. Start with a page change, a sales-call excerpt, or a question about a competitor. Get a sourced brief, a proposed battlecard edit, a pattern check, an internal update, or a cited answer for a rep.
Each skill works from supplied evidence. You can try the examples without a monitoring account, CRM, or Slack connection. The skills run in Claude Code or Codex; connectors and scheduling are optional additions.
In Claude Code:
/plugin marketplace add edoc33/ci-stack-skills
/plugin install ci-stack@ci-stack
Follow the install summary. Run /reload-plugins if it asks you to.
The commands follow Claude Code's plugin installation.
Or from a terminal (the installation path tested for this release):
claude plugin marketplace add edoc33/ci-stack-skills
claude plugin install ci-stack@ci-stackFor Codex, run these commands in a terminal:
git clone https://github.com/edoc33/ci-stack-skills.git
cd ci-stack-skills
python3 scripts/install_codex.py --dry-run
python3 scripts/install_codex.pyIf you already have the repository, use the upgrade instructions below instead of cloning again. You can also download the ZIP, extract it, and run the two Python commands from the extracted folder.
The local installer copies all eight skills and their required reference files to ~/.agents/skills.
Use --dest /your/skills/path for another location, or --skill battleagent to install only the new
skill. It preserves existing skills unless you explicitly use --update, which backs them up first.
Restart Codex if the new skills do not appear. Codex supports this
user skill location.
Python 3.10 or later is required only for the installer and repository checks.
Already installed? Follow the upgrade instructions to get version 0.5.0 with backups of existing local customizations.
After installing, paste one line into your agent:
Claude Code:
/ci-stack:battleagent Run the bundled demo inline using only its supplied evidence. Do not browse, write files, or send anything.
Codex:
$battleagent Run the bundled demo inline using only its supplied evidence. Do not browse, write files, or send anything.
You get a rep's question, evaluated against shared guidance and dated evidence, with a cited answer, unknowns, and the next discovery question. The fictional demo includes an outdated card so you can see how the skill handles a claim that needs review.
For real work, supply your question plus a permitted card, source excerpt, or shared knowledge location. A database is optional. See how to connect shared knowledge.
In Claude Code, type /ci-stack:ci-triage, a space, then the prompt below in the same message.
In Codex, use $ci-triage with the prompt below:
This is a fictional example. Return the brief here; no files or live tools needed.
We sell to IT teams that require SSO. Our current comparison says AcmeFlow only
includes SSO on Enterprise. We need to review that claim before a sales call.
Source: https://acmeflow.example/pricing, US page, signed out.
Before, captured 2026-08-04: "Team: SSO available on Enterprise only."
After, captured 2026-08-06: "Team: SSO included."
These pasted extracts are the preserved evidence. We have no product test.
Does this change our seller guidance? Leave unknown owners and deadlines unknown.
You should get a draft that distinguishes the published change from proof that SSO works. It should name the evidence, uncertainty, and recommended next action. It should leave approval pending. A usable result may recommend checking the claim before editing the battlecard.
| Your question | Skill | What you get |
|---|---|---|
| What should we monitor? | ci-portfolio |
A decision-linked watchlist, ignore rules, alert prompts, and a generic URL export |
| Does this change matter? | ci-triage |
A decision brief with evidence, uncertainty, and a recommended response |
| Can we support this claim? | ci-corroborate |
A source comparison, verification verdict, and supportable draft wording |
| Which battlecard lines need changing? | battlecard-patch |
A proposed diff against the current card, including stale claims and reversals |
| What can our rep conclude for this deal? | battleagent |
A concise answer from shared guidance and dated evidence, with citations, gaps, and the next question |
| Is this one deal or a recurring issue? | ci-pattern-check |
A count over a defined cohort, coverage limits, and a scoped conclusion |
| Who needs this reviewed guidance? | battlecard-patch in deal-brief mode |
An owner-specific draft linked to approved guidance and a verified active evaluation |
| What should the team know this period? | ci-weekly |
A weekly, twice-monthly, monthly, or custom-period update and decision views |
| What did the buyer say about a competitor? | ci-call-mentions |
A contextual Slack draft or a proposed CRM tag matched to an exact opportunity |
Start with the skill that fits the input you already have. There is no required sequence through all eight. For a new CI program, the usual loop is portfolio → triage → period update.
- First runs for all eight skills
- How the workshop workflows fit together
- Where Clef and Clairic could help (optional assessment; no model integration required)
- Evidence and file contract
- Validation and known limits
- 27 source examples, with dated captures and limits
- PMA workshop materials
The workshop branch retains dated exercises and reference outputs. Install from this repository's
main branch for the current skills and use the first-run guide above for current setup.
Tell the agent your decision, customer segment, known competitors, and where to save CI work.
It can reuse these details in ci-context.md under that directory. Start with the
fictional context example. Include owners and deadlines
when known. Missing metadata stays unknown; it should not prevent an initial draft.
Use permitted, minimized exports for internal calls and CRM data. Your agent provider processes what you supply under your account and organization settings. This repository adds no telemetry, background service, or credentials. Keep private outputs out of public repositories.
An AI skill provides analysis instructions. A connector retrieves or writes data. A runner such as n8n or GitHub Actions starts and coordinates the job. Your CRM and second brain supply deal context, battlecards, brand rules, and history. Slack or Teams carry the reviewed output.
The repository includes skill instructions, examples, and a local Codex installer. It does not
include deployed n8n workflows, GitHub Actions jobs, CRM field adapters, a timeline database, or a
newsletter delivery service. ci-call-mentions documents optional authorized Slack delivery; it
requires a connector and a durable delivery record. The other skills produce drafts and proposals.
See the workflow guide for the pieces each example needs.
The eight workshop workflows map to the original seven skills. The new battleagent adds a
rep-question workflow over the same evidence. Optional output modes cover
CRM tag proposals, competitor timelines, affected-deal briefs, and review-based discovery questions.
These modes create local drafts from supplied evidence; they do not add live integrations.
Created by Eric Do Couto, Head of Marketing at Visualping, for the PMA competitive intelligence workshop. Visualping is one optional source of page-change evidence. This is a personal, MIT-licensed repository. Issues and pull requests are welcome.
Run python3 -m unittest discover -s tests -v before proposing installer changes. For skill changes,
include a realistic input and the observed output behavior. See LICENSE.