Stop baking absolute mise path into validation file cmd field by leopoldjoy · Pull Request #705 · base/contract-deployments

leopoldjoy · 2026-05-28T16:04:39Z

Problem

PR #704 wrapped toolchain invocations with $(MISE_EXEC), which resolves at Makefile-parse time to either:

mise exec -- — when mise is on PATH, or
$(HOME)/.local/bin/mise exec -- — fallback when it is not.

The fallback case bakes a machine-specific absolute path into the cmd field of every committed validation JSON via:

Multisig.mk's GEN_VALIDATION macro (used by most current tasks), and
setup-templates/template-set-bridge-partner-threshold/Makefile's gen-validation target (which writes validations/signer.json via state-diff).

Both write a forge command into a JSON that is committed to git and hashed by the task-signing-tool at verification time. A validation file generated on machine A (with $HOME=/Users/alice) will fail signature verification on machine B (with $HOME=/Users/bob). This was discovered while finalising mainnet/2026-05-27-update-aggregate-verifier — three task-origin signatures had to be regenerated because the committed validation files contained the original signer's home directory.

The execution macros MULTISIG_APPROVE and MULTISIG_EXECUTE, and the inline $(MISE_EXEC) usages in other templates (template-gas-increase, template-pause-bridge-base, template-pause-superchain-config, template-upgrade-fault-proofs, template-safe-management), are unaffected because their $(MISE_EXEC) is only used inside the make recipe shell — nothing is written to disk.

Fix

Multisig.mk: in GEN_VALIDATION's embedded --forge-cmd, replace $(MISE_EXEC) with the bare literal mise exec --. The generated JSON is now portable across machines. Added a NOTE comment explaining the trade-off.
setup-templates/template-set-bridge-partner-threshold/Makefile: same fix in the template's inline state-diff invocation.
Makefile: extend bootstrap-mise with a command -v mise check that prints a clear warning + remediation when mise is installed at $HOME/.local/bin/mise but is not on the user's PATH. The signer-tool re-executes the validation file's cmd in a fresh shell that does not inherit the Makefile's $(MISE) resolution, so mise must be discoverable via PATH for the re-execution to work.
README.md: document the PATH requirement in the existing "Toolchain (mise)" section.

Why keep `mise exec` at all (i.e. why not strip it entirely)?

mise exec is what isolates this repo's pinned Foundry/Node/Go versions from any system-level foundryup or global toolchain installs. Stripping it would risk version drift between signers. The pattern we want is: the validation file carries the machine-agnostic token mise exec --, and each signer's environment ensures mise is resolvable.

Verification

make -n bootstrap-mise parses cleanly.
The signer-tool subprocess sees mise exec -- forge script ... in the generated cmd, with no absolute paths.
Other templates that use $(MISE_EXEC) only inside recipe shells (not embedded in JSON output) are intentionally left unchanged.

Follow-ups

This will be the basis for the two upcoming testnet tasks (Sepolia + Zeronet aggregate-verifier hash updates). Both will inherit the fixed GEN_VALIDATION behaviour from main once this lands.

PR #704 wraps toolchain calls with $(MISE_EXEC), which resolves at Makefile-parse time to either 'mise exec --' (when mise is on PATH) or the absolute fallback '$(HOME)/.local/bin/mise exec --' (when it is not). The execution macros above GEN_VALIDATION (MULTISIG_APPROVE, MULTISIG_EXECUTE) only use that expansion inside the make recipe shell, so the absolute fallback is harmless. GEN_VALIDATION is different: the embedded --forge-cmd string is written verbatim into the validation JSON's 'cmd' field, which is then committed to git and re-hashed at verification time on other machines. Baking an absolute, user-specific path into that file made committed validation files non-portable and caused signature-verification failures whenever the verifier's $HOME differed from the signer's. Fix: - Multisig.mk: replace $(MISE_EXEC) with the bare literal 'mise exec --' inside GEN_VALIDATION's --forge-cmd. The generated JSON is now portable. - Makefile: extend bootstrap-mise with a PATH check that prints a clear warning + remediation when 'mise' is installed but not on PATH, since the signer-tool re-execs the cmd in a fresh shell that does not inherit the Makefile's $(MISE) resolution. - README.md: document the PATH requirement in the existing Toolchain section.

cb-heimdall · 2026-05-28T16:04:46Z

✅ Heimdall Review Status

Requirement Status More Info

Reviews

✅ 2/2

Denominator calculation

Show calculation

1 if user is bot 0

1 if user is external 0

2 if repo is sensitive 0

From .codeflow.yml 2

Additional review requirements

Show calculation

Max	0

0

From CODEOWNERS 0

Global minimum 0

Max 2

2

1 if commit is unverified 0

Sum 2

The template's gen-validation target uses state-diff to write validations/signer.json, embedding a forge command that contained $(MISE_EXEC). Same bug as Multisig.mk's GEN_VALIDATION: the absolute path resolved at Makefile-parse time gets baked into the committed JSON, breaking signature verification on other machines. Replace with the bare literal 'mise exec --'.

jackchuma

How about we just exclude mise exec from gen validation commands altogether? We can trust that the internal engineers running these commands have mise working

cb-heimdall · 2026-05-28T16:45:27Z

Review Error for jackchuma @ 2026-05-28 16:45:27 UTC
User failed mfa authentication, see go/mfa-help

leopoldjoy · 2026-05-28T16:47:36Z

How about we just exclude mise exec from gen validation commands altogether? We can trust that the internal engineers running these commands have mise working

Agreed! But doesn't this new solution also work on that assumption?

Two validation files generated by the signer-tool for the nested PROXY_ADMIN_OWNER -> {CB_MULTISIG, BASE_SECURITY_COUNCIL} multisig: - base-signer.json: CB Coordinator Safe approve path - base-signer-2.json: Mock OP / Mock Security Council approve path Post-processing applied to each file: - Stripped absolute mise path from 'cmd' field ($HOME-specific /Users/.../.local/bin/mise -> bare 'mise exec --'), keeping the committed JSON portable across signer machines. Same fix as PR #705 applies upstream to Multisig.mk. - Filled in placeholder descriptions: * PROXY_ADMIN_OWNER approvedHashes slot: 'Approves the transaction hash for execution by the nested multisig.' * DGF gameImpls[621] slot: 'Updates the game type 621 implementation in the DisputeGameFactory from the old AggregateVerifier to the new AggregateVerifier with updated TEE_IMAGE_HASH, ZK_RANGE_HASH, and ZK_AGGREGATE_HASH.' - Disabled task-origin validation: removed 'taskOriginConfig' block, added 'skipTaskOriginValidation: true' (this task does not ship task-origin signatures).

Two validation files generated by the signer-tool for the nested PROXY_ADMIN_OWNER -> {CB_MULTISIG, BASE_SECURITY_COUNCIL} multisig: - coinbase-signer.json: CB Signer Safe approve path - security-council-signer.json: Security Council Safe approve path Filename naming follows the mainnet pattern (signer-tool renders it as the role label) rather than the May-18 testnet precedent's 'base-signer{,-2}.json'. Makefile gen-validation-* targets and FACILITATOR.md updated to match. Post-processing applied to each file: - Stripped absolute mise path from 'cmd' field ($HOME-specific /Users/.../.local/bin/mise -> bare 'mise exec --'), keeping the committed JSON portable across signer machines. Same fix as PR #705 applies upstream to Multisig.mk. - Filled in placeholder descriptions: * DGF gameImpls[621] slot: 'Updates the game type 621 implementation in the DisputeGameFactory from the old AggregateVerifier to the new AggregateVerifier with updated TEE_IMAGE_HASH, ZK_RANGE_HASH, and ZK_AGGREGATE_HASH.' * PROXY_ADMIN_OWNER approvedHashes slot: 'Approves the transaction hash for execution by the nested multisig.' - Disabled task-origin validation: removed 'taskOriginConfig' block, added 'skipTaskOriginValidation: true' (this task does not ship task-origin signatures).

* Sepolia: upgrade ZK and TEE hash (2026-05-28) Mirrors mainnet/2026-05-27-update-aggregate-verifier on sepolia so the testnet exercises the same TEE image + ZK range program ahead of mainnet rollout. Follows the existing sepolia/2026-05-18-upgrade-zk-and-tee-hash template verbatim; only the three hash values in .env change: - TEE_IMAGE_HASH: updated to mainnet target - ZK_RANGE_HASH: updated to mainnet target - ZK_AGGREGATE_HASH: unchanged from previous testnet deployment (set explicitly for clarity; mainnet target equals it) All other AggregateVerifier immutables (DELAYED_WETH, TEE_VERIFIER, ZK_VERIFIER, ANCHOR_STATE_REGISTRY, CONFIG_HASH, BLOCK_INTERVAL, INTERMEDIATE_BLOCK_INTERVAL, L2_CHAIN_ID, GAME_TYPE) are read from the existing on-chain AggregateVerifier at deploy/upgrade time, so no env plumbing is required for them. Task does not ship task-origin signatures; FACILITATOR.md instructs the facilitator to set 'skipTaskOriginValidation: true' on each generated validation file before committing. * Sepolia: commit AggregateVerifier deploy artefacts Deployed new AggregateVerifier at 0xeCe9c5b9DCa09f1a0Ed85DA97f8F1396dC5634Ce on sepolia in tx 0xaef9dc24...07ac4 (block 10941556, Etherscan-verified). - addresses.json: aggregateVerifier populated - README.md: Transactions section added with explorer + artefact links - records/DeployAggregateVerifier.s.sol/11155111/run-1779990011323.json: forge-broadcast artefact * Sepolia: add validation files (task-origin validation disabled) Two validation files generated by the signer-tool for the nested PROXY_ADMIN_OWNER -> {CB_MULTISIG, BASE_SECURITY_COUNCIL} multisig: - base-signer.json: CB Coordinator Safe approve path - base-signer-2.json: Mock OP / Mock Security Council approve path Post-processing applied to each file: - Stripped absolute mise path from 'cmd' field ($HOME-specific /Users/.../.local/bin/mise -> bare 'mise exec --'), keeping the committed JSON portable across signer machines. Same fix as PR #705 applies upstream to Multisig.mk. - Filled in placeholder descriptions: * PROXY_ADMIN_OWNER approvedHashes slot: 'Approves the transaction hash for execution by the nested multisig.' * DGF gameImpls[621] slot: 'Updates the game type 621 implementation in the DisputeGameFactory from the old AggregateVerifier to the new AggregateVerifier with updated TEE_IMAGE_HASH, ZK_RANGE_HASH, and ZK_AGGREGATE_HASH.' - Disabled task-origin validation: removed 'taskOriginConfig' block, added 'skipTaskOriginValidation: true' (this task does not ship task-origin signatures). * Sepolia: rename validation files for human-readable signer-tool labels base-signer.json -> coinbase-signer.json (CB Multisig path) base-signer-2.json -> security-council-signer.json (Base Security Council path) Filename is the role label shown in the signer-tool UI; 'base-signer-2' gives signers zero hint about which nested-safe path they are on. Mirrors mainnet/2026-05-27-update-aggregate-verifier naming convention and AGENTS.md guidance ('Always name the validation file(s) something simple ... [for] human readable names in the signer tool'). JSON content unchanged - signatures are computed over expectedDomainAndMessageHashes, not the filename - so no regeneration needed. Makefile gen-validation-* targets and FACILITATOR.md updated to match. * Sepolia: task executed — record approve + execute artefacts CB Multisig approval, Security Council approval, and final execution all succeeded on sepolia (chain 11155111): - CB approve: 0x33465e04d448f1627adfd6130fa2765bd672d0280a4aa190e79fd780d9d29d48 (block 10947347) - SC approve: 0x91f51a84ac2100ea2da72e9555ac423ef169f43239197a31a4d4e313b1f77838 (block 10947374) - Execute: 0xe6b2fde0ac66fc17e5ecb6b56aaa0eddbc567130ad66a9dff2c25376320f8e32 (block 10947378) README status flipped from READY TO SIGN to [EXECUTED](https://sepolia.etherscan.io/tx/0xe6b2fde0...32) and the Transactions section extended with explorer + artefact links for the two approvals and the execution.

leopoldjoy requested a review from jackchuma May 28, 2026 16:07

0x00101010 approved these changes May 28, 2026

View reviewed changes

leopoldjoy requested a review from 0x00101010 May 28, 2026 16:16

0x00101010 approved these changes May 28, 2026

View reviewed changes

leopoldjoy mentioned this pull request May 28, 2026

Sepolia: upgrade ZK and TEE hash (2026-05-28) #706

Merged

jackchuma requested changes May 28, 2026

View reviewed changes

leopoldjoy mentioned this pull request May 28, 2026

Zeronet: upgrade ZK / TEE hash, bump anchor root (2026-05-28) #707

Open

jackchuma approved these changes May 28, 2026

View reviewed changes

leopoldjoy merged commit 0b13931 into main May 28, 2026
14 checks passed

leopoldjoy deleted the leopold/fix-multisig-mk-absolute-mise-path branch May 28, 2026 17:03

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stop baking absolute mise path into validation file cmd field#705

Stop baking absolute mise path into validation file cmd field#705
leopoldjoy merged 2 commits into
mainfrom
leopold/fix-multisig-mk-absolute-mise-path

leopoldjoy commented May 28, 2026 •

edited

Loading

Uh oh!

cb-heimdall commented May 28, 2026 •

edited

Loading

Uh oh!

jackchuma left a comment

Uh oh!

cb-heimdall commented May 28, 2026

Uh oh!

leopoldjoy commented May 28, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

leopoldjoy commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Fix

Why keep mise exec at all (i.e. why not strip it entirely)?

Verification

Follow-ups

Uh oh!

cb-heimdall commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ Heimdall Review Status

Uh oh!

jackchuma left a comment

Choose a reason for hiding this comment

Uh oh!

cb-heimdall commented May 28, 2026

Uh oh!

leopoldjoy commented May 28, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

leopoldjoy commented May 28, 2026 •

edited

Loading

Why keep `mise exec` at all (i.e. why not strip it entirely)?

cb-heimdall commented May 28, 2026 •

edited

Loading