Chaining Plugins

Individual plugins are powerful. Chained plugins are transformative. The sales plugin researches a prospect. The proposal plugin drafts a proposal using that research. The legal plugin reviews the proposal for problematic language. Three plugins, one pipeline, and the output is a reviewed proposal built on real research — not a generic template.

The chain works because each plugin is a domain specialist. The sales plugin knows what matters in prospect research: company size, tech stack, recent news, decision-maker profiles, competitive landscape. The proposal plugin knows how to structure a consulting proposal: problem statement, approach, timeline, investment, team. The legal plugin knows what clauses to flag: indemnification, scope limitations, payment terms, termination conditions. No single plugin has all three knowledge domains. The chain combines them.

The practical execution uses prompt queuing. You submit the research prompt, and before it finishes, you queue the proposal prompt that references the research output files. Before the proposal finishes, you queue the legal review prompt that references the proposal output. Co-Work chains them automatically — each stage executes the moment the previous stage completes.

The critical detail is file references. Each stage must save its output to a named file, and the next stage must reference that file explicitly. "Read the prospect-research.md file and use it as the basis for this proposal." "Read the proposal-draft.md file and review it for legal concerns." File names are the API of your pipeline. If you rely on conversation context instead of files, compaction will destroy your intermediate work.

1. 1. Map the Plugin Chain Identify which plugins participate and in what order. Each plugin stage should produce a specific output file that the next stage consumes. Draw the chain on paper: Plugin A produces file X, Plugin B consumes file X and produces file Y.
2. 2. Queue the Full Pipeline Submit all prompts in sequence using prompt queuing. Each prompt explicitly references the file produced by the previous stage. Co-Work chains execution automatically.
3. 3. Review at Checkpoints Not every stage needs a human checkpoint. But the transition from research to drafting is a good place to review — bad research produces bad proposals regardless of how good the proposal plugin is. Decide in advance which transitions need human review.
4. 4. Save the Chain as a Skill Once you have a working plugin chain, encode it as a skill with parameterized inputs. "Prospect Pipeline: given a company name, research → propose → review." The skill fires the full pipeline from a single prompt.