--prompt-template and ~/.prime/agent/prompts/ are not reaching system context in --print mode

[Propershare]

Summary

When using prime-agent --print with --prompt-template <path> or with files auto-discovered from ~/.prime/agent/prompts/, the template content does not reach the model as system context. The model behaves as if no prompt template was supplied.

To reproduce

# 1. Install prime-agent v0.7.1
PRIME_AGENT_PACKAGE=prime-agent sh install.sh

# 2. Wire Ollama (any model — reproduced with gemma4:e4b and gemma4:26b)
mkdir -p ~/.prime/agent
cat > ~/.prime/agent/models.json <<'JSON'
{
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434/v1",
      "api": "openai-completions",
      "apiKey": "ollama",
      "compat": { "supportsDeveloperRole": false, "supportsReasoningEffort": false },
      "models": [{ "id": "gemma4:e4b" }, { "id": "gemma4:26b" }]
    }
  }
}
JSON

# 3. Drop a prompt template
mkdir -p ~/.prime/agent/prompts
cat > ~/.prime/agent/prompts/00-test.md <<'MD'
---
name: test
description: Test template
---
The secret word is BANANA.
MD

# 4. Try to use it
prime-agent --print --provider ollama --model gemma4:e4b --no-session \
  "What is the secret word mentioned in your prompt template?"

Expected: the model answers BANANA (or a paraphrase of the template).

Actual: the model says it has no context, asks for the document, etc.

I was unable to retrieve the information... Could you please provide the document, resource identifier, or section name associated with this?

What I tried

1. --prompt-template /home/suspect/.prime/agent/prompts/00-test.md (full file path) — same result.
2. --prompt-template /home/suspect/.prime/agent/prompts (directory) — same result.
3. Inlining the same content into the user message works perfectly. So the template loader is producing the right content; the issue is that the loaded template is never appended to the system prompt (or print mode doesn't read it from the right place).

Environment

• prime-agent v0.7.1
• macOS / Linux (Ubuntu 22, kernel-venv from auto-bootstrap)
• Provider: Ollama (openai-completions compat)
• Mode: --print (also relevant for --mode json / --mode rpc)

Workaround

For headless runs, inline the template content into the user message. For interactive TUI, the template seems to land correctly (per the docs), but I haven't yet isolated a single repro inside the TUI.

Likely cause (from a quick source read)

loadPromptTemplates in dist/core/prompt-templates.js is wired correctly (it discovers agentDir/prompts/ when includeDefaults: true), and resource-loader.js calls it — but in --print mode the loaded templates do not appear to be appended to the system prompt. The expandPromptTemplate helper exists for /prompt slash command expansion, which suggests prompt templates may only be visible to /prompt, not to the system prompt in print mode.

Happy to send a PR if you can point at the right place to wire this.

[Hotragn]

I think this is a usage mismatch rather than a --print bug, and there is a real docs gap behind it. From reading the source on main:

Prompt templates are invocations, not injected context. docs/prompt-templates.md opens with "Prompt templates are Markdown snippets that expand into full prompts. Type /name in the editor to invoke a template, where name is the filename without .md." Nothing loads a template into the system prompt — discovery only registers /name, and the body is substituted into the user message when you invoke it. So 00-test.md registers /test, and asking "what is the secret word mentioned in your prompt template" never touches it. That also explains your step 3: inlining the content works because inlining is what invocation does.

For your repro, this should expand:

prime-agent --print --provider ollama --model gemma4:e4b --no-session "/test"

Expansion is not disabled in print mode as far as I can tell — expandPromptTemplates defaults to true for ordinary prompts in AgentSession, and I could not find a print/RPC path that overrides it to false. If /test also comes back with no context on your setup, then there is a real bug and it is worth reporting that specific command, because it is a different failure from the one described here.

If you want content that is always in context, templates are the wrong mechanism. Two supported options:

• ~/.prime/agent/AGENTS.md (or CLAUDE.md), loaded automatically at startup and appended as project context. This is the closest match to what you were expecting.
• --append-system-prompt <text>, repeatable, for per-invocation additions.

--verbose lists the prompt templates that were actually discovered, which is the quickest way to confirm --prompt-template resolved your path before debugging anything further.

I have no affiliation with the project and have not run your exact Ollama setup — this is from tracing the loader and the submission-normalization path, so please sanity-check it against your box.

[kevinjosethomas]

Hi, thanks for taking the time to report this! Since open sourcing Prime Agent, we’ve received far more Issues than we can reliably triage and maintain. Rather than leave a large backlog that we cannot meaningfully work through, we’re closing the current Issue queue and moving to a discussion-first process.

We have established new contribution guidelines to help us continue iterating on Prime Agent and better manage contributions from the community. Going forward, GitHub Issues will be our maintainer work queue rather than the initial place for public reports and requests. Bug reports, feature requests, questions, and other feedback should begin in GitHub Discussions. Please read the full process documented in our contribution guidelines. Security vulnerabilities should continue to be reported privately according to our security policy.

While we’re closing this backlog, we’re still reviewing it at a high level to identify recurring bugs, useful ideas, and important problems that we should address ourselves. Thanks again for the time you put into this!