Observed Behavior
After running teamwork init or /setup-teamwork, every agent file receives an identical Project Knowledge section populated with the same tech stack facts: languages, build command, test framework, package manager, lint command.
The Tester receives the build command. The Documenter receives the test framework. The Security Auditor receives the package manager. The Architect receives the lint command. None of these facts are relevant to those roles.
This is confirmed by reviewing .github/skills/setup-teamwork/SKILL.md (Step 4: Fill In Placeholders), which applies the same detected values uniformly across all agent files.
Why It Matters
Context that is irrelevant to a role wastes tokens and trains agents to ignore the Project Knowledge section entirely. Worse, it crowds out context that is relevant — if the Tester's Project Knowledge section is full of build and lint commands, the actually critical facts (test command, coverage tool, test directory structure) may be buried or absent.
Role-specific knowledge also enables better MCP tool selection: the Security Auditor who knows the auth mechanism and sensitive paths can use Semgrep MCP more effectively; the Architect who knows the ORM and migration tool can use the ADR MCP more effectively.
v2 Target
Each agent template should define a role-specific knowledge schema using typed placeholder comments:
<!-- KNOWLEDGE:tester: test_framework, test_command, coverage_tool, test_dir -->
<!-- KNOWLEDGE:coder: languages, build_command, lint_command, package_manager -->
<!-- KNOWLEDGE:security-auditor: auth_mechanism, sensitive_paths, dependency_audit_command -->
<!-- KNOWLEDGE:architect: orm, migration_tool, api_framework, database_engine -->
The setup-teamwork skill and teamwork init should parse these schemas and populate each agent file with only the facts it needs. Facts not relevant to a role are omitted entirely.
Observed Behavior
After running
teamwork initor/setup-teamwork, every agent file receives an identical Project Knowledge section populated with the same tech stack facts: languages, build command, test framework, package manager, lint command.The Tester receives the build command. The Documenter receives the test framework. The Security Auditor receives the package manager. The Architect receives the lint command. None of these facts are relevant to those roles.
This is confirmed by reviewing
.github/skills/setup-teamwork/SKILL.md(Step 4: Fill In Placeholders), which applies the same detected values uniformly across all agent files.Why It Matters
Context that is irrelevant to a role wastes tokens and trains agents to ignore the Project Knowledge section entirely. Worse, it crowds out context that is relevant — if the Tester's Project Knowledge section is full of build and lint commands, the actually critical facts (test command, coverage tool, test directory structure) may be buried or absent.
Role-specific knowledge also enables better MCP tool selection: the Security Auditor who knows the auth mechanism and sensitive paths can use Semgrep MCP more effectively; the Architect who knows the ORM and migration tool can use the ADR MCP more effectively.
v2 Target
Each agent template should define a role-specific knowledge schema using typed placeholder comments:
The
setup-teamworkskill andteamwork initshould parse these schemas and populate each agent file with only the facts it needs. Facts not relevant to a role are omitted entirely.