my-hermes-skills

Has oh-my-pi adopted Pi 1.0 and Pi Durable, and should it?

Abstract

Background. Earendil released Pi 1.0 and the experimental Pi Durable package on 2026-10-01[1]. oh-my-pi (omp), a fork of Pi, would gain upstream features or durability from adopting either, at some integration cost.

Objective. To determine whether can1357/oh-my-pi has adopted Pi 1.0 or Pi Durable, and whether it should.

Methods. The Earendil announcement was read, and omp's releases, commits, code, pull requests and issues were queried through gh (GitHub search and API). Nothing was measured locally.

Results. omp is a hard fork on its own version line (v18.11.0, released 2026-10-14)[2]. Its only link to Pi 1.x is a compatibility shim extended by PR #14909 so that upstream Pi 1.x extensions can load[3]. A code search for pi-durable returns no files. omp's durability work is native session-persistence hardening plus two open proposals[4, 5].

Conclusions. No: omp has adopted neither Pi 1.0 nor Pi Durable. It should not rebase onto Pi 1.0, but it should keep closing Pi 1.x extension-compatibility gaps and track Pi Durable's design for its own durability proposals.

1. Introduction

Pi is Earendil's coding agent; omp (can1357/oh-my-pi) forked from it and now releases on its own schedule. Pi 1.0 and Pi Durable shipped together on 2026-10-01. Pi Durable is a separate harness framework of about 15k lines, with SQLite, JSONL and in-memory storage; it covers long-running and multi-surface agents and is explicitly "not a replacement" for the Pi coding agent[1]. Whether omp picks these up affects which extensions its users can run and how durable its sessions are.

2. Methods

2.1 Sources and search strategy

The Earendil "Pi Durable" post[1] was fetched on 2026-10-15. On the same date, omp's releases[2], PR #14909[3], issues[4, 5], porting doc[6], PR #12699[7], and code and commits in the repository[8] were queried via gh (GitHub search + API). Search terms included pi-durable (code search) and durability keywords (commit search); the full query list was not recorded.

2.2 Evidence classification

2.3 Experimental setup

Not applicable; nothing was measured locally.

2.4 Analysis

Each candidate adoption path was checked for direct evidence (dependency, merged code, open PR or issue) in omp's repository.

3. Results

3.1 Adoption status (RQ1, RQ2)

Table 1. omp adoption status as of 2026-10-15 (repository records).

Itemomp statusEvidence
Pi 1.0 core (rebase / dependency)Not adoptedReleases are v18.x (v18.11.0, 2026-10-14)[2]. The fork keeps @oh-my-pi/ packages and maps @earendil-works/ imports onto them (docs/porting-from-pi-mono.md)[6].
Pi 1.x extension compatibilityPartial, shim onlyPR #14909 (merged 2026-10-10)[3] ports upstream pi 1.0.2/1.1.0 root exports (getCurrentSystemMessage/transcript utils, compositeTuiLine, parseSkillBlock) into the legacy shim. The PR lists gaps that are still open.
Pi Durable (@earendil-works/pi-durable)[1]Not adoptedCode search for pi-durable returns no files. Commit hits are unrelated session-durability fixes.
Durable-harness ideas, done nativelyProposals onlyIssue #14386 (durable connection recovery and automatic task continuation, OPEN, 2026-10-06)[4]. Issue #6947 (session-durable workflow DAGs, OPEN)[5].
Session crash durability (omp's own)Shipped, ongoingFor example, PR #11560 (stale rewrites no longer erase durable turns, 2026-09-15) and 6c84a8bb (completed entries survive crashes).

4. Discussion

4.1 Principal findings

RQ1 and RQ2: omp has not rebased onto Pi 1.0 and does not use Pi Durable. omp is still a hard fork with its own version line[2], treating upstream Pi as a source of features to port, not as a base to track. Because Pi Durable is a separate package and not part of the coding agent, omp would have to integrate it on purpose, and nothing suggests that is planned.

4.2 Comparison with prior work

Earendil positions Pi 1.0 as a stable foundation and Pi Durable as experimental[1]. omp's existing session durability (#11560, 6c84a8bb) and its open proposals #14386 and #6947[4, 5] address overlapping goals natively rather than through Pi Durable.

4.3 Practical implications

RQ3: omp should keep porting selectively rather than rebase onto Pi 1.0. It should not take Pi Durable into its core now, but it could support Pi Durable as an optional add-on.

Table 2. Adoption recommendations (interpretation).

ChangeRecommendationReasoning
Rebase onto Pi 1.0Noomp has diverged into its own v18 line with its own packages, Rust natives, metaharness and auth gateway (repo layout, release notes)[8, 2]. A rebase would mean rewriting most of omp. Its existing approach (porting doc, cherry-picks like #12699)[6, 7] already pulls in upstream improvements piece by piece.
Pi 1.x extension compatibilityYes, keep closing the gapsPi 1.0 is described as a stable foundation (Earendil post)[1], so its extension API should change less often from now on. Covering it in the shim gives omp users the upstream extension ecosystem cheaply. PR #14909 shows the pattern works and lists the gaps still open[3].
Pi Durable as omp's core runtimeNot nowEarendil labels Pi Durable experimental and says it "does not replace" the coding agent[1]. omp already has its own session durability (#11560, 6c84a8bb), and moving onto an experimental runtime would put at risk crash-safety work omp has already shipped.
Pi Durable ideas / optional integrationYes, worth trackingPi Durable's features (storage backends, reconnecting from different surfaces, remote execution environments, multiple people steering one agent)[1] match omp's open proposals #14386 and #6947[4, 5]. omp could borrow the design, or let Pi Durable host omp-compatible tools through an adapter. It shares pi-ai with omp, so integration is plausible (ASSUMPTION: the API surfaces are still compatible enough to adapt; not verified).

The main risk of not adopting is that Pi 1.x and omp keep drifting apart, so each new upstream feature costs more to port. Keeping the compatibility shim test-pinned (as #14909 does)[3] keeps that cost visible and contained.

5. Limitations

6. Conclusion

omp has adopted neither Pi 1.0 nor Pi Durable; it remains a hard fork that ports upstream features selectively and bridges Pi 1.x extensions through a shim. The sensible path is to keep extending that shim, not rebase, and to borrow from Pi Durable's design for omp's own durability proposals rather than adopt the experimental runtime.

References

  1. [1] Pi Durable (Earendil Engineering, 2026-10-01): https://earendil.com/posts/pi-durable/ back to text 1 2 3 4 5 6 7 8 9 10 11
  2. [2] oh-my-pi v18.11.0 release (GitHub, 2026-10-14): https://github.com/can1357/oh-my-pi/releases/tag/v18.11.0 back to text 1 2 3 4 5 6
  3. [3] PR #14909, bridge upstream pi 1.x shim gaps (can1357/oh-my-pi): https://github.com/can1357/oh-my-pi/pull/14909 back to text 1 2 3 4 5 6 7
  4. [4] Issue #14386, durable connection recovery proposal (can1357/oh-my-pi): https://github.com/can1357/oh-my-pi/issues/14386 back to text 1 2 3 4 5 6 7
  5. [5] Issue #6947, session-durable workflow DAGs (can1357/oh-my-pi): https://github.com/can1357/oh-my-pi/issues/6947 back to text 1 2 3 4 5 6
  6. [6] docs/porting-from-pi-mono.md (can1357/oh-my-pi): https://github.com/can1357/oh-my-pi/blob/main/docs/porting-from-pi-mono.md back to text 1 2 3 4
  7. [7] PR #12699, port upstream pi cache-warming (can1357/oh-my-pi): https://github.com/can1357/oh-my-pi/pull/12699 back to text 1 2 3 4
  8. [8] oh-my-pi repository (GitHub): https://github.com/can1357/oh-my-pi back to text 1 2 3