Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
46 commits
Select commit Hold shift + click to select a range
234af16
feat(skills): add independent code cleanup workflow
adamwhp Oct 1, 2026
56da523
refactor(write-skill): adopt portable authoring guidance and validation
adamwhp Oct 1, 2026
6383581
refactor(skills): streamline foundation and elicitation workflows
adamwhp Oct 1, 2026
f7a6768
refactor(skills): clarify engineering guidance and nesting measurement
adamwhp Oct 1, 2026
1517aa2
refactor(skills): clarify operational workflows and report validation
adamwhp Oct 1, 2026
f8d3c04
docs(skills): clarify authoring guidance and record refactor evidence
adamwhp Oct 1, 2026
1c8a1e6
refactor: retire obsolete catalogue metadata and artefacts
adamwhp Oct 2, 2026
40331ce
fix(skills): align workflows with repository guidance
adamwhp Oct 2, 2026
c9aa204
refactor(skills): standardise template content placeholders
adamwhp Oct 2, 2026
1489af0
Update readme
adamwhp Oct 2, 2026
5d14d60
fix(skills): keep pull request descriptions succinct
adamwhp Oct 2, 2026
d1f773a
chore: remove maintenance artefacts
adamwhp Oct 5, 2026
f142bc6
feat(skills): expand prose guidance with paired examples
adamwhp Oct 5, 2026
1c36feb
docs(write-prose): simplify review introduction and opening example
adamwhp Oct 5, 2026
2d5f46d
refactor(skills): compose write-prose for human-facing artefacts
adamwhp Oct 5, 2026
4f10ce1
fix(elicit): restore relentless decision tree traversal
adamwhp Oct 5, 2026
b2ddaf1
refactor(write-skill): define concise authoring contracts and checks
adamwhp Oct 5, 2026
856f40a
refactor(skills): remove implicit invocation settings
adamwhp Oct 5, 2026
387723b
test: reset suite to write-skill helpers
adamwhp Oct 5, 2026
3d43fcd
refactor(write-skill): consolidate validation and token counting
adamwhp Oct 5, 2026
03d567f
refactor(elicit): sharpen challenge and clarification
adamwhp Oct 5, 2026
72d0f6d
refactor(write-skill): apply prose and description conventions
adamwhp Oct 5, 2026
87c2556
refactor(skills): strengthen continuous domain context maintenance
adamwhp Oct 6, 2026
daa6427
refactor(write-prose): clarify writing for humans and agents
adamwhp Oct 6, 2026
664b539
refactor(skills): simplify commit and pull-request workflows
adamwhp Oct 6, 2026
0fa46cf
refactor(maintain-agents): focus on concise project instructions
adamwhp Oct 6, 2026
fe5cfe4
chore(skills): remove review-architecture skill
adamwhp Oct 6, 2026
4207736
refactor(skills): streamline implement and debug workflows
adamwhp Oct 6, 2026
f5d308d
refactor(skills): simplify modular design guidance
adamwhp Oct 6, 2026
6886a2f
docs(skills): standardise reference filenames
adamwhp Oct 6, 2026
e0ddb1b
docs(skills): record lowercase reference paths
adamwhp Oct 6, 2026
bff7fd0
refactor(measure-code-complexity)!: measure current code with bundled…
adamwhp Oct 6, 2026
b24ca62
refactor(code-cleanup)!: isolate review and expand deletion guidance
adamwhp Oct 7, 2026
7cf52a1
refactor(test-design): protect behavioural promises through supported…
adamwhp Oct 7, 2026
16b22b1
docs(test-design): use lowercase reference filename
adamwhp Oct 7, 2026
e28261c
refactor(test-design)!: remove the obsolete TDD skill
adamwhp Oct 7, 2026
b9478ef
refactor(code-review)!: isolate reviewers and save consolidated reports
adamwhp Oct 7, 2026
1ce9e94
refactor(research): delegate investigation and simplify reports
adamwhp Oct 7, 2026
792fc21
refactor(skills): make input requirements direct
adamwhp Oct 7, 2026
5b315f7
refactor(upgrade-dependencies): make upgrade verification explicit
adamwhp Oct 7, 2026
b32fb4e
refactor(review-pull-request): delegate review to code-review
adamwhp Oct 7, 2026
77adfff
refactor(skills): refine product planning workflow
adamwhp Oct 7, 2026
c39a37b
refactor(skills): delegate procedures to their owning skills
adamwhp Oct 7, 2026
0f12a9f
style(debug): align references with prose conventions
adamwhp Oct 7, 2026
6676589
refactor(debug): enforce systematic investigation before repair
adamwhp Oct 7, 2026
16fff4f
docs: simplify lifecycle guidance and add a coherent example
adamwhp Oct 7, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .gitignore
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
.DS_Store
.idea
/node_modules
node_modules/
/docs
2 changes: 1 addition & 1 deletion .oxfmtrc.json
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
{
"$schema": "./node_modules/oxfmt/configuration_schema.json",
"ignorePatterns": ["node_modules/**", ".opencode/node_modules/**"],
"ignorePatterns": ["**/node_modules/**"],
"printWidth": 80,
"proseWrap": "preserve",
"tabWidth": 4,
Expand Down
2 changes: 1 addition & 1 deletion .oxlintrc.json
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@
"correctness": "error",
"suspicious": "error"
},
"ignorePatterns": ["node_modules/**", ".opencode/node_modules/**"],
"ignorePatterns": ["**/node_modules/**"],
"options": {
"denyWarnings": true,
"reportUnusedDisableDirectives": "error",
Expand Down
8 changes: 3 additions & 5 deletions AGENTS.md
Original file line number Diff line number Diff line change
@@ -1,7 +1,5 @@
- When a user correction establishes a reusable repository-wide rule, ask whether to invoke `$maintain-agents` to add it.
- When authoring or reviewing a skill, include only runtime instructions and knowledge for its agent; exclude authoring rationale, evaluation history, and composition discussion, and justify behaviour only with runtime-available context.
- When eliciting within a software project, invoke `$elicit-with-context` instead of `$elicit`; use `$elicit` directly only for non-software work or as part of `$elicit-with-context`.
- When composing skills, rely on each invoked skill's public contract; do not restate or steer its internal logic.
- Default to a user-agent workflow; introduce team roles, named accountable people, approval bodies, or third-party coordination only when the user or project evidence requires them.
- When a user correction establishes a durable project-wide instruction needed in every session, ask whether to invoke `$maintain-agents` to add it.
- When creating, reviewing or maintaining skills, invoke `$write-skill`.
- Default to a user-agent workflow; do not introduce team roles, named accountable people, approval bodies, or third-party coordination.
- Assume the execution harness supports fresh agents; do not add fallback handling for their unavailability.
- After implementing changes run `bun run checks` before handoff.
32 changes: 8 additions & 24 deletions GLOSSARY.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,38 +2,22 @@

## Language

### Skill model
### Skill invocation

**Predictability**: The degree to which a skill makes the agent follow the same process on every run, without requiring the same output.
**Lifecycle entry point**: A skill a user invokes to begin a common software delivery task, such as product definition, ticket planning, implementation or maintenance. Users can start at the entry point that matches their available context and desired outcome.

**User-invoked skill**: A skill the user selects explicitly. This is Propulsion's default invocation mode.
**Supporting skill**: A skill that supplies a focused capability to a lifecycle entry point through composition. Users can also invoke supporting skills directly when they need that capability.

**Model-invoked skill**: A skill the agent may select autonomously or invoke from another skill. Use this exception only when autonomous discovery would naturally help during ordinary coding work often enough to earn its permanent context cost.
**Breadth-first planning**: Establish product scope before detailing individual features, then break a selected feature into small executable tickets. Each level narrows the work without making earlier levels mandatory for an understood task.

**Invocation policy**: Client-specific metadata within a skill bundle that controls whether an agent may select that skill implicitly. The same intent may require different policy fields in different clients.
**Explicit invocation**: A skill is selected by name in the request. This is Propulsion's default entry mode.

### Skill types
**Implicit invocation**: The agent selects a skill without a named request because its trigger applies.

**Skill type**: The authority category recorded in a skill's frontmatter as `metadata.type`. Propulsion uses exactly `performer`, `router`, `teaching`, or `utility`.
**Composed invocation**: A skill invokes another skill as part of its workflow, using the called skill's public contract. Composition is distinct from implicit selection.

**Performer skill**: Owns one bounded requested outcome end to end, including its actions, artefacts, verification, handoff, and stopping boundary.

**Router skill**: Resolves the inputs needed to coordinate other skills and returns their results. It contains only routing unique to that outcome; called skills remain authoritative and the router neither repeats nor overrides their context.

**Teaching skill**: Supplies knowledge and decision rules inside a caller's workflow. It does not own an independent operation, mutation, artefact, verification, handoff, or stopping decision.

**Utility skill**: Owns one bounded reusable operation and returns its result, whether the operation runs in the calling agent, a fresh agent, or a deterministic tool.
**Duplicated procedure**: A composing skill or supporting resource repeats execution rules owned by a skill it delegates to. Caller-specific inputs, sequencing, authority boundaries and outcome checks define composition rather than duplicated procedure.

### Software quality

**Deterministic measurement**: A reproducible tool-produced observation of one named property of a fixed software candidate, with its scope, components, configuration, limitations, and comparison basis preserved. An attention threshold requires appraisal but is not by itself a defect, an overall quality score, or permission to weaken behaviour or design.

### Skill benchmarking

**Authentic workflow arm**: One isolated benchmark run that gives a skill set the same implementation task, starting repository, model, tools, permissions, and environment as the other arms while allowing it to load and follow its own required skills and workflow.

**Adaptive single-run comparison**: One fresh run per benchmark arm, followed by another run only for an arm whose result is tied, failed, or surprising enough that run-to-run variation could materially change the conclusion.

**Skill-context tokens**: The exact tokenizer count of a benchmark arm's skill discovery metadata and every skill or reference file loaded during the run, reported separately from whole-run model usage.

**Runtime token usage**: The execution harness's complete input, cached-input, output, and reasoning-output token counts for one benchmark arm, excluding the independent quality evaluator.
Loading
Loading