Quick fix
# One-liner: run the conflict precheck first (see write/port surface declarations)
站内文档 --check
Install as needed; no forced order.
What you will run into
You want to build a complete Agent toolchain (orchestration + skills + memory + gateway + security + review + observability) and are not sure whether the tools can be installed together or will step on each other's config. Conclusion: 17 tools, 57 measured pairs - all compatible.
Why this happens
The 17 Agent-toolchain tools have 57 measured-compatible pairs in total:
| Type | Pairs | Notes |
|---|---|---|
| No shared write surface | 51 pairs | Each has its own CLI/directory; naturally independent |
Shared ~/.claude.json write surface | 6 pairs | agt/deepcode/code-review-graph/claude-mem related pairs; write different keys/sections, measured coexistence without overwrites |
Of the 210 fully declared pairs in the conflict matrix, this entry only includes the measured verified ones; the remaining inferred states (17 possible-conflict pairs) wait until measured before being recorded - no measurement, no claim.
Whether multiple tools can coexist is decided by "overlap of write surfaces": a tool's write surface is the set of files/directories it exclusively reads and writes at runtime (config files, state dirs, caches).
Tool pairs with no shared write surface coexist naturally (they never perceive each other); when they share one surface (e.g.
~/.claude.json), coexistence stays safe as long as each writes its own keys/sections and the format merges by key (JSON/ini-like) — conflicts only occur when two tools write the same location with different semantics or overwrite wholesale.
Deterministic takeaway: to judge whether two tools conflict, first list their write surfaces (config paths, state dirs) and compare; the 57 verified-compatible pairs mean these 17 tools' write surfaces are pairwise disjoint;
"not tested, not claimed" is not an excuse — the 17 possible-conflict pairs must be tested before becoming verified, because write-surface analysis only rules out obvious conflicts, it does not prove the absence of conflicts.
Step by step
1. Conflict precheck — Required
站内文档 --check
First check the write/port surface declarations; or inspect site docs.
2. Install as needed + smoke-test each tool — Required
After installing, run each tool's official CLI version command (exit 0); when installing skill-type tools (agent-skills/superpowers) together, verify the ~/.claude/skills/ directory listing.
3. For combined scenarios, see the combos — Required
Combos 001-003 (agt+crg mixed / crg+deepcode same venv / claude-mem+deepcode cross-ecosystem) all measured passing.
4. Re-smoke after upgrades — Required
Re-run smoke tests after major-version upgrades; if you hit a new conflict, submit a helpdesk card and it will be recorded after official re-measurement.
How to confirm it works
All per-tool smoke tests green (expected shape):
Each tool's official CLI version command -> all exit 0
ls ~/.claude/skills/ -> skill directory listing complete
Combo scenarios reproduce per 001-003 (measured).
Other environments
| Environment | Notes |
|---|---|
| 57 verified pairs | All carry tested_at (2026-08-27: 1 pair; 2026-08-28: 56 pairs) |
| 17 tools covered | agent-framework(16), agt(15), agent-skills(12), deepcode(9), code-review-graph(8) etc. |
| Measurement method | coexist-test.sh isolated HOME + venv, dual smoke exit 0 |
| Not measured | The 17 possible-conflict pairs are outside this entry's commitment; no measurement, no claim |
Symptom
You want to build a complete Agent toolchain (orchestration + skills + memory + gateway + security + review + observability) and are not sure whether the tools can be installed together or will step on each other's config.
Cause
The 17 Agent-toolchain tools have 57 measured-compatible pairs: 51 pairs share no write surface (each has its own CLI/directory, naturally independent); 6 pairs share the ~/.claude.json write surface (agt/deepcode/code-review-graph/claude-mem related pairs) but write different keys/sections, measured coexistence without overwrites. Of the 210 fully declared pairs in the conflict matrix, this entry only includes the measured verified ones; the remaining inferred states (17 possible-conflict pairs) wait until measured before being recorded.
Solution
- Install as needed, no forced order; precheck conflicts with
站内冲突预检说明(see write/port surface declarations) or inspect站内文档. - After installing, smoke-test each tool (official CLI version command exit 0); when installing skill-type tools (agent-skills/superpowers) together, verify the
~/.claude/skills/directory listing. - For combined scenarios, see combos 001-003 (agt+crg mixed / crg+deepcode same venv / claude-mem+deepcode cross-ecosystem), all measured passing.
- Re-smoke after major-version upgrades; if you hit a new conflict, submit a helpdesk card and it will be recorded after official re-measurement.
Verification Evidence
- All 57 verified pairs carry tested_at (2026-08-27: 1 pair; 2026-08-28: 56 pairs); details in site docs.
- 17 tools covered: agent-framework(16 pairs), agt(15), agent-skills(12), deepcode(9), code-review-graph(8), agentshield(8), swe-rex(7), logfire(6), herdr(6), claude-mem(6), superpowers(4), claude-code-router(4), oh-my-pi(4), omc(3), moltis(2), skillspector(2), hoop(2).
Local verification record
- The unmeasured part (17 possible-conflict pairs) is outside this entry's commitment, keeping "no measurement, no claim".
Applicable tools (17)
This problem is linked from the tool pages below — reachable from either side.
agent-frameworkagent-governance-toolkitagent-skillsagentshielddeepcodecode-review-graphclaude-code-routerlogfiremoltisoh-my-claudecodeherdrskillspectoroh-my-pihoopswe-rexsuperpowersclaude-memBundled solutions (combos)
This problem is part of the following verified bundles.
Source & evidence
← Multi-Agent Orchestration domain | Problem Catalog Experience | Facing another issue? Go to the Helpdesk