Menú
Use when Codex should call the local Research Sidecar app, create or inspect a Sidecar session, hand off explicit context, maintain a research graph, or use the Sidecar API/CLI for a second-opinion review.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Use when working in a research repository with graph.yaml, Markdown research notes, hypotheses, evidence, experiments, literature sources, or when Codex is asked to inspect, create, update, or maintain a lightweight research graph.
Use when a task is explicitly framed as research, scholarship, literature analysis, theory building, methodology critique, paper reading, academic writing, research notes, or evaluating a research intuition.
Use when the user explicitly asks to use Research Sidecar, open/start/connect to the sidecar app, inspect or select a sidecar research graph, install sidecar workspace skills, or run an external/second-opinion sidecar thinking pass.
Use when writing, revising, or generating explanatory research reports, experiment reports, analysis reports, 报告, 实验报告, Markdown reports, or HTML reports where the user needs to understand Codex-run work and make a decision from incomplete context.
Basado en la clasificación ocupacional SOC
| name | sidecar-thinking |
| description | Use when Codex should call the local Research Sidecar app, create or inspect a Sidecar session, hand off explicit context, maintain a research graph, or use the Sidecar API/CLI for a second-opinion review. |
Use this skill to hand off explicit context from Codex to the local Research Sidecar. The sidecar should judge only what is in the context packet; do not imply it has Codex's hidden state.
The app can be embedded inside a research repo or installed separately. If it is installed separately, run it with SIDECAR_WORKSPACE_ROOT=/path/to/research-repo.
Default to calling the sidecar and returning a session URL so the user can ask or continue in the web UI. Only use the fully automatic ask mode when the user explicitly wants Codex to send the question and relay the answer back.
cd sidecar && npm run devcd sidecar && npm run codex:call -- --title "Review" --context "..." --file SKILL.md --question "What is the weakest assumption?"--question in call mode adds a manual user message to the session without running the model. This is useful when Codex wants to stage the exact question while leaving the user in control of when to run or revise it.
If the user asks Codex to send the question and return the sidecar answer, use:
cd sidecar && npm run codex:ask -- --title "Review" --context "..." --file SKILL.md --question "What is the weakest assumption?"
Ask mode creates the same session and then streams the sidecar answer back to stdout. Treat it as a convenience path, not the default workflow.
The web UI supports workspace tools when the auto-selected API route is Chat Completions. Keep tools enabled when the sidecar should be able to list files, read workspace-relative files, inspect git diff, or load full workspace skill instructions itself. Use the CLI handoff files only for context that should remain stable and cache-friendly.
Workspace skills are discovered from SKILL.md frontmatter first. Strong matches are loaded into the sidecar context automatically; weaker candidates remain visible and can be loaded with the load_skill tool.
The graph is manifest-first:
graph.yaml owns nodes, edges, root, UI defaults, and file pointers.research/graph.yaml.SIDECAR_GRAPH_MANIFEST or .side/config.json at graph.manifestPath./ for explicit workspace-root-relative links.Useful API endpoints:
GET /api/research-graphGET /api/workspace/file?path=<path>GET /api/workspace/raw?path=<path>POST /api/sessionsPOST /api/sessions/:id/streamPrefer smaller, auditable packets over broad repository dumps. Attach only files that matter. If the sidecar needs more context, ask Codex for specific missing files or summaries.
Ask the sidecar to identify: