Content Repurposing Station
What it is
A workflow that turns approved notes, articles, and transcripts into reusable drafts under human publishing control.
Operating shape
- Difficulty:
starter - Cadence:
on demand - Category:
content
Core concepts
- Content Repurposing
- File Tools
- Skills
- Approval Boundaries
- Artifacts
Source evidence
vendor/hermes-agent/toolsets.py#<file>— Structural Source Map fallback for toolsets.py. A core source file in the Tool Registry and Toolsets structure within Tools, Toolsets, and MCP.
Related articles
- Seven Cost-Aware Hermes Workflows That Go Beyond Chat
- Ten Hermes Workflows That Turn a Chatbot into a 24/7 Assistant
Implementation pattern
Outcome
Transform one human-approved source into clearly labelled draft variants and a provenance manifest, without adding claims or publishing to any platform.
Before you start
- Use the synthetic
approved-source.mdfixture below first; replace it only with content whose owner has approved the named channels. - Use a disposable output folder. Disconnect publishing credentials and third-party social connectors.
- Define length and tone constraints for each destination, plus words or claims that require human review.
Input and output contract
Input is only approved-source.md and channel-rules.md. Output is drafts/linkedin.md, drafts/newsletter.md, drafts/short-posts.md, and drafts/manifest.md. Every factual claim must map to a source line or be labelled EDITORIAL SUGGESTION. The output must say DRAFT — NOT PUBLISHED, contain no invented quotes/statistics, and perform no network publishing.
| Draft file | Source lines used | New factual claims | Review status |
|---|---|---:|---|
| linkedin.md | 12-24 | 0 | pending human review |Build it
Create a complete, non-confidential fixture and its channel rules, then run a file-only bounded task:
LAB_DIR="$HOME/hermes-labs/content-repurposing"
mkdir -p "$LAB_DIR/drafts"
cat > "$LAB_DIR/approved-source.md" <<'EOF'
Owner: Hermes Lab Team
Approved: 2026-07-29
Allowed channels: LinkedIn, email newsletter, short social posts
Hermes can run scheduled tasks through its built-in cron system.
Script-only no-agent jobs run a reviewed script without invoking an LLM.
This lab prepares drafts for human review; it never publishes them.
EOF
cat > "$LAB_DIR/channel-rules.md" <<'EOF'
LinkedIn: 80-120 words; plain language; no hashtags.
Newsletter: 120-180 words; explanatory tone; one factual heading.
Short posts: exactly 3 variants, each at most 240 characters.
All channels: no new facts, quotes, statistics, testimonials, urgency, or calls to buy.
EOF
cd "$LAB_DIR"
hermes chat --toolsets "file" -q "Read only approved-source.md and channel-rules.md. Create drafts/linkedin.md, drafts/newsletter.md, drafts/short-posts.md, and drafts/manifest.md. Mark every draft 'DRAFT — NOT PUBLISHED'. Follow each channel rule, preserve factual meaning, cite source line ranges in the manifest, label any non-factual creative suggestion, invent no quote or statistic, make no network call, and do not publish or schedule content."Expected output is three readable channel drafts plus a manifest showing zero unsupported factual claims.
Verify it
- Diff every number, quote, name, and product claim against
approved-source.md. - Check each manifest line range and reject any draft whose source cannot be located.
- Search outputs for
DRAFT — NOT PUBLISHED; it must appear in every draft. - Confirm no publish/schedule command, URL request, or connector call occurred.
- Replace the source with a two-sentence fixture, empty
drafts/, and rerun in a fresh session; no old claims may leak into the new outputs.
If it fails
- If a claim lacks provenance, delete it or label it as a non-factual editorial suggestion before review.
- If channel tone overwhelms the source meaning, shorten the brief and require sentence-level fidelity.
- If old content appears, start a fresh session and empty disposable output folder.
- If any connector tries to publish, revoke its credential, stop the run, and return to file-only mode.
Safety, privacy, and cost
Only process content you own or are authorized to reuse; preserve attribution and licensing requirements. Remove private customer details before model use. Human review is mandatory for legal, medical, financial, brand, and promotional claims. One source and three bounded drafts control model cost; avoid autonomous retry loops. Platform publishing is intentionally not part of this built-in Hermes file workflow.