Skip to content

chore(release): merge release/0.1.0 into main - #51

Merged
a1exk-dev merged 35 commits into
mainfrom
release/0.1.0
Oct 18, 2026
Merged

a1exk-dev merged 35 commits into
mainfrom
release/0.1.0

Conversation

@a1exk-dev

Copy link
Copy Markdown
Owner

Summary

First release. Brings 9 merged PRs from develop into main (#1, #18, #19, #29, #30, #31, #32, #33, #50).

sequenceDiagram
    participant Human
    participant main
    participant release.yml
    participant develop
    Human->>main: merge release/0.1.0
    main->>release.yml: pull_request closed (merged)
    release.yml->>release.yml: release job: tag v0.1.0, GitHub release
    release.yml->>develop: back-merge job: open PR main -> develop, auto-merge
Loading

Evidence

  • Before: no v* tag on main; release.yml not on main.
    After: check the release run: tag v0.1.0, release v0.1.0, and the back-merge PR into develop merged by auto-merge.

Merge Danger

Door: one-way

Users install skills from main, and the tag v0.1.0 is public once made.

Blast Radius: repo-wide

All skills, tooling, and CI on develop reach main.

Refs: #49, #40

Add a Knowledge rule to AGENTS.md: prototypes/ holds throwaway work,
so agents take facts and decisions from the rest of the repo and open
prototypes/ only when the human's current prompt asks for it.
Add a Knowledge rule to AGENTS.md that keeps agents out of prototypes/ unless the human's current prompt asks for it.
Record the glossary and memory entries from the agent-instructions
skill spec map: the Agent instruction term, Rule placement,
ASD-STE100 and ISO content limits, target platforms, portable
frontmatter and pointers, and test file placement.

Refs: #2
Add the agent-instructions Skill under skills/writing/ with its
reference files, and its eval suite under tests/skills/writing/.

Beyond the spec, the Skill warns about the install folder, saves one
fallback report per message when no subagent is available, checks the
word count of changes after the last review round, and rewords the
base-list definitions for rule 1.11.

Refs: #17
The ASD-STE100 reviewer saves its sentence list to a temporary file and
returns only the path and the findings. Each reader gives only the 3
required parts, one line per step.

Refs: #17
Add the agent-instructions Skill under skills/writing/ with its
reference files and its eval suite under tests/skills/writing/.

Refs: #17
Replace the 500-line warning with a 20,000-byte error (about 5,000
tokens). Claude Code keeps only the first 5,000 tokens of a skill after
compaction, and Codex cuts the middle of a long shell read. Record the
limit and its reasons in MEMORY.md.

Refs: #28
Apply the accepted findings of the round 1 review. Declare the
skill's bold terms as technical nouns, replace non-approved words,
put the report steps in a numbered list, and remove text that
repeats the reference files. Set the SKILL.md limit in
loading-rules.md to 20,000 bytes, to match the repo check. Let a
word in capital letters name itself without counting as emphasis.

Rename "Rule file" to "Rule", the CONTEXT.md term, and record in
MEMORY.md that a Skill term follows the CONTEXT.md term.

Refs: #28
Apply the accepted round 1 review findings to the
agent-instructions Skill, and rename "Rule file" to
the CONTEXT.md term "Rule".

Refs: #28
…uctions

Cover all 53 ASD-STE100 Issue 9 rules and GR-1 to GR-8 in
ste-rules.md, in our own words with rule numbers, and correct the
words of the Skill. Limit sentences by text type, use CAUTION only
for data loss or a change that cannot be undone, and define the
report "warning" apart from the STE signal word WARNING.

Run the reader test with 3 tasks for each changed file, and correct
the text after each round. Give each reference file a scope and a
pointer to SKILL.md, settle the open reader questions, and write the
metadata and block scalar rules without new nouns.

Run the eval matrix until each statement passes with the skill
(115/115). Give tests 01 and 10 the release facts, so that no run
waits for a user. Make the no-subagent checks save one report for
each message, and copy the end-of-run report without changes.

Update eval statements [1.3], [5], and [7] and the grader notes for
Statements 6 and 7.

Refs: #17
Apply the ASD-STE100 Issue 9 book check to the
agent-instructions Skill, and run the eval matrix until
each statement passes with the skill.

Refs: #17
- Rename CONTEXT.md to GLOSSARY.md and rewrite it in the
  domain-modeling format, because the installed mattpocock/skills
  read only GLOSSARY.md.
- Make the agent-instructions skill accept GLOSSARY.md and CONTEXT.md.
- Route architecture decisions to ADRs in docs/adr/, written only
  after approval; keep other durable guidance in MEMORY.md.
- Add the Agent skills block to AGENTS.md and update
  docs/agents/issue-tracker.md from the current template.
- Track .agents/skills/ and the .claude/skills/ symlinks, so every
  clone gets the same agent skills.
- Ignore .claude/worktrees/, which holds agent worktrees.
- Add the git-commit skill draft under prototypes/.
Replace the typos and bash check steps with setup-bun, a frozen
install, and bun run check. Delete scripts/check.sh and _typos.toml.
Point the AGENTS.md Checks line to the container check and fix
commands, add a README Setup section, and point the MEMORY.md
SKILL.md size entry to scripts/check.ts.

Refs: #47
release.yml now makes every v<semver> tag.

Refs: #44
@a1exk-dev
a1exk-dev merged commit fd16846 into main Oct 18, 2026
6 checks passed
@a1exk-dev
a1exk-dev deleted the release/0.1.0 branch October 18, 2026 20:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant