Synrepo

name: synrepo description: Use synrepo in repositories with a .synrepo/ directory. Prefer synrepo cards, compact search, and bounded task contexts before reading source files cold.

synrepo context

synrepo is a local, deterministic code-context compiler. It compiles repository files into observed graph facts, code artifacts, task contexts, and cards/MCP packets before agents read source cold.

Codex setup

This skill belongs at .agents/skills/synrepo/SKILL.md.

Project-scoped setup writes trusted project .codex/config.toml. Global Codex MCP registration is not automated by synrepo setup; if you configure ~/.codex/config.toml manually, launch synrepo mcp and pass repo_root to repo-addressable tools. To add local non-blocking Codex nudges, run synrepo setup codex --agent-hooks.

For project-scoped manual setup, edit trusted project .codex/config.toml directly:

[mcp_servers.synrepo]
command = "synrepo"
args = ["mcp", "--repo", "."]

Agent doctrine

synrepo is a local, deterministic code-context compiler: repo files -> graph facts -> code artifacts -> task contexts -> cards/MCP. In .synrepo/ repos, prefer MCP/CLI over cold reads for questions, reviews, search, impact, and edits.

Default path

For questions, reviews, search routing, and edits: orient, ask or search, cards, impact or risks, edit, tests, changed.

1. Start with synrepo_orient before reading the repo cold.
2. Use synrepo_ask for broad plain-language tasks needing one bounded, cited task-context packet.
3. Use synrepo_find only for bounded file-routing suggestions after broad task context is clear. Use synrepo_search for exact files, symbols, strings, flags, code-shaped errors, and tool names. For broad lexical searches, prefer output_mode: "compact".
4. Use tiny cards to route and normal cards to understand. Use synrepo_minimum_context once a focal target is known and neighborhood risk is unclear.
5. Use synrepo_impact (or synrepo_risks) before risky edits or reviews, and synrepo_tests before claiming done.
6. Use synrepo_changed after edits to review changed context and validation commands.
7. After stale resumes or lost context, call synrepo_resume_context before asking the user to repeat repo state.
8. Read full source files or request deep cards only after bounded cards identify the target or prove insufficient.

MCP repository selection

Project-scoped MCP configs launching synrepo mcp --repo . have a default repository; omit repo_root or pass the absolute root when known.

Global MCP configs that launch synrepo mcp serve registered projects by absolute path. In global or defaultless contexts, pass the workspace absolute path as repo_root. If a tool reports an unmanaged repository, ask the user to run synrepo project add <path>; do not bypass registry gating.

Graph-backed facts are authoritative. Overlay commentary, explain docs, and proposed cross-links are advisory and freshness-sensitive. Existing explain reads are safe when useful: use synrepo_explain with budget=deep for 1-3 focal targets and synrepo_docs_search for architecture/why questions. Stale labels are information. Refresh is explicit: fresh commentary requires synrepo mcp --allow-overlay-writes and synrepo_refresh_commentary(target).

Client-side hooks for Codex and Claude may nudge before direct grep, read, review, or edit workflows and emit [SYNREPO_CONTEXT_FAST_PATH], [SYNREPO_DETERMINISTIC_EDIT_CANDIDATE] Intent: ..., or [SYNREPO_LLM_NOT_REQUIRED]. Hooks are advisory; source mutation still requires synrepo mcp --allow-source-edits and synrepo_apply_anchor_edits.

Do not

• Do not open large files first. Start at tiny and escalate only when a specific field forces it.
• Do not read a full source file before synrepo routing identifies it; full-file reads are explicit escalation.
• Do not treat overlay commentary, explain docs, or proposed cross-links as canonical source truth. They are advisory prose layered on structural cards.
• Do not generate or refresh explain (--generate-cross-links, commentary generate/refresh) unless the task justifies the cost; cached explain reads are allowed.
• Do not ask the user to repeat stale repo context until synrepo_resume_context has been tried.
• Do not expect watch or background behavior unless synrepo watch is explicitly running.
• Do not mistake client-side hook nudges for MCP enforcement.

Product boundary

• synrepo stores code facts and bounded operational memory. It is not a task tracker, not session memory, and not cross-session agent memory.
• synrepo_resume_context is an advisory repo packet regenerated from existing state. It is not prompt logging, chat history, raw tool-output capture, or generic session memory.
• Handoff or next-action lists are derived recommendations regenerated from repo state. External systems own assignment, status, and collaboration.
• Freshness is explicit. A stale label is information, not an error; it is not silently refreshed on read.

MCP tools (primary interface)

• synrepo_readiness() - cheap read-only preflight for graph, overlay, index, watch, reconcile, and edit-mode status
• synrepo_orient() - workflow step 1: small routing summary before reading the repo cold
• synrepo_ask(ask, scope?, shape?, ground?, budget?) - default high-level front door for one bounded, cited task-context packet
• synrepo_search(query, literal?, limit?, output_mode?, budget_tokens?) - exact lexical search for symbols, flags, code strings, schema keys, file paths, and validation
• synrepo_explain(target, budget?) - bounded card lookup; use budget=deep for 1-3 focal targets when existing overlay commentary would help
• synrepo_docs_search(query, limit?) - advisory search over existing materialized explain docs for architecture, intent, gotchas, and why questions
• synrepo_card(target?, targets?, budget?, budget_tokens?) - structured card for one file or symbol, or a small batch
• synrepo_context_pack(goal?, targets?, budget?, budget_tokens?, output_mode?, include_tests?, include_notes?, limit?) - batch known read-only code artifacts and task-context pieces into one token-accounted response
• synrepo_task_route(task, path?) - cheap route classification when only intent, budget, and next tools are needed
• synrepo_minimum_context(target, budget?) - bounded neighborhood once a focal target is known
• synrepo_impact(target) or synrepo_risks(target) - first-pass change-risk context before edits or risky reviews
• synrepo_tests(scope) - discover likely validation commands before claiming done
• synrepo_changed() - review changed context and validation guidance after edits
• synrepo_resume_context(limit?, since_days?, budget_tokens?, include_notes?) - compact repo resume packet before asking the user to repeat stale context
• synrepo_overview() - full dashboard only when the full operational picture is useful

synrepo_ask returns answer, cards_used, evidence, grounding, omitted_context_notes, next_best_tools, and context_packet. Its grounding policy accepts mode or citations, include_spans, and allow_overlay; default to observed graph/index evidence and allow overlay only when advisory machine-authored context is acceptable.

Graph facts are authoritative observed source truth. Overlay commentary, explain docs, and notes are advisory; LLM-authored output never mutates the canonical graph. Embeddings are optional routing/search helpers and are not the core trust source.

Node IDs: file_0000000000000042, symbol_0000000000000024. Use synrepo_search to find them.

CLI fallback

synrepo status                                   # health check
synrepo status --recent                          # bounded operational history
synrepo orient                                  # small routing summary
synrepo ask "review this module"                 # bounded task-context packet
synrepo resume-context --json                    # compact repo resume packet
synrepo task-route "find auth entrypoints"        # advisory route classifier
synrepo search <query>                           # lexical search
synrepo node <target>                             # node metadata as JSON (accepts paths, symbol names, or node IDs)
synrepo graph query "inbound <target>"            # reverse dependencies
synrepo graph query "outbound <target>"           # forward dependencies
synrepo graph stats                              # counts by type
synrepo reconcile                                # refresh graph
synrepo links list                               # cross-link candidates
synrepo findings [--freshness <state>]           # findings summary